1
INSTRUMENTATION AND COMPUTER CONTROL SYSTEMS SENSORS AND SIGNAL CONDITIONING
Steve Collins
Michaelmas Term 2012
Introduction
An instrumentation system obtains data about a physical system either for the purpose of collecting information about that physical system or for the feedback control of the physical system
Any instrumentation system must include an input transducer (sensor), such as a strain gauge, whose response to a particular stimulus can be
measured electrically. The other component that is generally present in modern instrumentation systems is a digital processor, such as a computer or a micro-controller. These programmable components have
the flexibility to be used for a variety of functions. The most important
function that they perform is to convert data into information. In the
simplest situation the processing required to extract information may only
involve converting an input signal by a scale factor so that the final result is
in conventional units. For example, the output voltage signal from a strain
gauge may be converted to the corresponding actual strain. Alternatively,
within a more sophisticated system the signal from a strain gauge placed
on an engine mounting might be processed to extract the vibrational
spectrum of an engine, which is then used to detect any unusual frequency
that might be indicative of wear. This information can then be displayed to a
2
user, stored for later analysis, transmitted to a remote location or used by a
controller.
The signal from a transducer is usually analogue in nature, ie. it is
continuously varying and can take any value (within an allowed range).
This continuous analogue data has to be converted to a digital format prior to being transferred to the digital processor. Any instrumentation system must therefore include an analogue-to-digital (A/D) converter (ADC for short) to convert an analogue signal into a digital format, such as those discussed in the first year P2 course.
A typical ADC will be an existing component that has been designed to
convert an analogue input voltage, typically with a range of a few volts, into a digital word, which usually contains 8 or more bits. However, the output from a typical transducer, such as a strain-gauge, might have an amplitude of less than 10 mV. This transducer
output signal must therefore be amplified in an analogue signal conditioning circuit before it can be converted into a digital word.
Another aspect of the performance of the ADC that must also be taken into
consideration when designing the signal conditioning circuit is that the ADC
samples the transducer output at specific time intervals. An unfortunate consequence of this is that several frequencies will become indistinguishable at the ADC output. This is referred to as aliasing, and the effect can only be avoided by using a low-pass, anti-alias filter to ensure that only the low frequencies that can be represented
accurately are present in the signal applied to the ADC input. Since the
3
requirement for the anti-alias filter arises from a fundamental property of
the ADC, this type of filter should always be present.
Figure 1: A block diagram of a typical instrumentation
system with several different output devices
As shown in Figure 1 the characteristics of typical sensors and ADCs mean
that the data collection (or acquisition) part of a typical modern
instrumentation system can be split into the three functional blocks, a
sensor, signal conditioning circuits and an ADC. The digital output from the
ADC can then be processed in a programmable digital processor to extract information that can be displayed to an operator, stored in a memory or transmitted via a data link or used in feedback control.
The costs of all the components are continually falling. It is therefore
becoming economically viable to gather an increasing amount of data, and
hence hopefully information, from an every expanding range of host
systems. One example of this trend is a 2.25 Km suspension bridge
4
constructed for the 2000 Olympics that had approximately 300 sensors
embedded within the structure. These include:
strain gauges to keep track of framework fatigue
sensors to monitor motion in the stay cables caused by cross winds
accelerometers in the roadway to measure the impact of earthquakes
The data from these sensors are gathered by four separate data-
acquisition units (one in each pier of the bridge). These linked units are
then connected to offices at the bridge site, in the headquarters of the
bridge operating company in Athens and in the headquarters of the
structural monitoring division of one of the bridge builders, which is in
France.
This technology trend and its impact on every conceivable system means
that all engineers should be familiar with instrumentation systems.
5
Aims and Organisation of the Course
The aim of the sensors and signal conditioning course is to develop an
understanding of the function of the first key parts of a typical
instrumentation system, such as the one in Figure 1.
The first parts of the instrumentation system that will be considered are the
sensors. There are many different sensors that rely upon one of a range of
different physical phenomena to create an output signal in response to
different stimuli. A comprehensive survey of all sensors is time-consuming
and beyond the scope of this course. However, we will aim to give a brief
survey (or list) of sensor types and the types of signals which might be
produced. Such signals are typically rather weak. These signals must
therefore be amplified before they are converted into digital words.
One problem caused by the small amplitude of the output signals from sensors is that they can be easily confused with other small voltage changes within the instrumentation system. Techniques to
reduce the interference caused by these other small voltage changes,
including careful design of the circuit layout, shielding it from external
electromagnetic fields and creating a signal represented by the voltage
difference between two signals, will be briefly described. The resulting
small differential signals could be amplified by a differential amplifier
containing a single operational amplifier (op-amp). However, this circuit is
not ideal and the more complex, but easier to use, instrumentation
6
amplifier, is therefore used. This was discussed in the first year P2 course,
but it is so important we will cover it again here.
Once the analogue output signals have been amplified they need to be converted into a digital word. Any instrumentation system must
therefore include an analogue to digital converter (ADC). Two types of
ADCs that are often used in instrumentation systems were discussed in the
first year P2 course. The flash converter is conceptually simple and fast.
However, it necessarily contains a large number of components and it is
therefore relatively expensive. In some situations it is necessary to use an
alternative type of converter. The alternative converter that was described
in the P2 course, known as the successive approximation converter,
contains a digital to analogue converter (DAC). These DACs are also
useful in digital control systems, such as the one shown in Figure 2. In this
course, the discussion of ADCs will concentrate on issues relating to
interfacing.
Figure 2: A block diagram of a typical digitally based control system.
7
Amplification of the sensor output signal is designed to match the maximum expected sensor output signal to the maximum input signal of the ADC. The sensitivity of the resulting data is then determined by the minimum change in the signal that can be reliably detected. The
performance of ADCs means that this important aspect of the system
performance is often limited by undesirable signals which are generated
within the components of the electronic circuit. The origin of these noise
signals, their effect on instrumentation systems and methods to limit their
effects will therefore be described.
The discussion of noise highlights the fact that one of the important system parameters that determines the amount of noise in a system is the system bandwidth. Filter circuits that can be used to control the bandwidth of the system therefore play a critical role in limiting the impact of noise on a system. In fact there are four key different types of
filters that are commonly used in instrumentation systems to fulfil a variety
of functions. The circuits and characteristics of such filters were discussed
in the first year P2 course. In this course the characteristics, applications
and implementation of each of various filters will be further described.
Amplifiers and filters within an instrumentation system are typically based upon op-amps. The function of a particular circuit within one of these systems can be understood by analysising the circuit with the assumption that the op-amp is ideal. However, all real op-amps have a
finite input impedance, gain and output impedance. One of the key stages of designing any op-amp based circuit is the selection of the
8
particular op-amp that should be used so that it 'appears' to be ideal in the particular circuit that is being designed. Even after an op-amp
has been selected that appears to be ideal additional components may be
required in a circuit to compensate for other non-ideal aspects of an op-
amp’s behaviour. Finally, all op-amps have a gain that reduces at high frequencies. This means that any op-amp will only appear to be ideal for frequencies less than a maximum value.
The output voltage from a strain gauge and several other sensors is a d.c.
voltage. However, there are sensors in which the stimulus of interest causes a change in either capacitance or inductance. Changes in these two parameters can only be sensed if an a.c. signal is applied to the sensor. In addition, to avoid strong sources of interference, a.c. signals
can also be applied to circuits containing sensors such as strain gauges.
The use of this approach, with a lock-in amplifier, will be discussed.
9
Syllabus and Learning Outcomes Sensors and signal conditioning. Interference avoidance and
instrumentation amplifiers. Non-ideal op-amps. Sources of noise (including
quantisation noise) and noise reduction by bandwidth limitation; Filters and
their applications.
At the end of this course students should be able to: 1. An appreciation of the importance of signal conditioning for the
interfacing of sensors.
2. An understanding of the key types of signal conditioning: amplification,
filtering and isolation.
3. An understanding of interference, and the roles of differential and
instrumentation amplifiers.
4. An appreciation of the impact of “real” operational amplifiers, and an
understanding of how engineers can allow for real op-amp parameters in
circuit analysis.
5. An appreciation of the origins of noise in signal conditioning circuits, and
how its impact can be estimated.
6. An understanding and knowledge of basic filter types, together with their
implementation.
7. An understanding of the importance of bandwidth limitation using filters
(including anti-aliasing) and the lock-in amplifier.
10
Textbooks
A textbook that covers the majority of the topics in this course at an
appropriate level is
Design with operational amplifiers and analog integrated circuits
by Sergio Franco, published by McGraw-Hill
Additional material on noise and the design of low-noise systems is
contained in
Low-Noise Electronic System Design
by C.D. Motchenbacher and J.A. Connelly
published by John Wiley and Sons.
For further reading on the subject of sensors
Sensors and Transducer: Characteristics, Applications,
Instrumentation and Interfacing
by M.J. Usher and D.A. Keating,
published by MacMillan Press Ltd
or
'Instrumentation for Engineers and Scientists'
by John Turner and Martyn Hill,
published by the OUP.
11
AN INTRODUCTION TO SENSORS The quantities measured by instrumentation systems are almost invariably
non-electrical; for example, pressure, displacement, temperature, etc...
The first step in any electronic system that gathers data on this type of quantity must be to find a device that will transform a change in the physical quantity of interest into an electrical signal. This
transformation occurs in a device called a transducer; thus, within a
platinum resistance thermometer a change in temperature is converted into
a change in resistance using the temperature dependence of the resistance
of the platinum wire.
A transducer may be described as an input transducer (now more usually known as a sensor) or an output transducer (now more usually known as an actuator), depending on the direction of information
flow. Examples of input transducers are thermometers, microphones,
pressure sensors and photodiodes; the corresponding output transducers
are heaters, loudspeakers, pistons and light-emitting diodes.
There are many different types of sensor for some physical quantities (for
example, temperature, strain, light flux etc...). In addition, there are other
physical quantities, such as pressure and viscosity, which can only be
measured if a mechanical transducer is used to convert the primary
variable, such as a pressure, into a secondary mechanical variable, such
as strain in a thin membrane, which can be measured. A description of the
various sensors available to measure each physical quantity could be the
12
subject of an entire course of lectures. This type of review is not within the
scope of this course, but we discuss some key issues in relation to sensors
and some typical sensors.
Some key sensor parameters A sensor's sensitivity indicates how much the sensor's output changes
when the measured quantity changes. As a simple example, if a platinum
resistance thermometer changes resistance by 0.4 ohm when the
temperature changes by 1 °C, the sensitivity is 0.4 ohm/°C. Sensors that
measure very small changes must have very high sensitivities. Sensors
also have an impact on what they measure; for instance, a room
temperature platinum resistance thermometer inserted into a hot liquid
cools the liquid while the liquid heats the thermometer. Sensors need to be
designed to have a small effect on what is measured. Making the sensor
smaller often improves this and may introduce other advantages.
Technological progress allows more and more sensors to be manufactured
on a microscopic scale, such as microsensors using MEMS technology. In
most cases, a microsensor reaches a significantly higher speed and
sensitivity compared with macroscopic approaches.
A good sensor obeys the following rules:
• Is sensitive to the measured property
• Is insensitive to any other property
• Does not influence the measured property
13
Ideal sensors are designed to be linear. The output signal of such a sensor
is linearly proportional to the value of the measured property. The
sensitivity is then defined as the ratio between output signal and measured
property. For example, if a sensor measures temperature and has a
voltage output, the sensitivity is a constant with the unit [V/K]; this sensor is
linear because the ratio is constant at all points of measurement. In general
it is rather difficult to design sensors which are linear over wide ranges. For
example, for the platinum resistance thermometer the resistance as a
function of temperature can actually be expressed as:
( )[ ]1001 320 −+++= TCTBTATRRT
(see http://en.wikipedia.org/wiki/Resistance_thermometer)
where the quadratic and cubic terms are small, but not necessarily
negligible. It might be possible to correct for this non-linearity in the signal
conditioning stage. However the programmability of microcontrollers and
microprocessors means that it is easier to perform this correction after
analogue to digital conversion.
Sensor deviations More generally, if the sensor is not ideal, several types of deviation can be
considered:
• The sensitivity may in practice differ from the value specified. This is
termed a sensitivity error, but the sensor may still be linear.
• Since the range of the output signal is always limited, either by the
voltages powering any circuits or by the ADC input range, the output
14
signal will eventually reach a minimum or maximum when the measured
property exceeds the limits – often referred to as saturation. The full-
scale range defines the maximum and minimum values of the measured
property.
• If the output signal is not zero when the measured property is zero, the
sensor has an offset or bias. This is defined as the output of the sensor
at zero input – it is very common.
• If the sensitivity is not constant over the range of the sensor, this is
termed nonlinearity. Usually this is defined by the amount the output
differs from ideal behaviour over the full range of the sensor.
• If deviation is caused by a rapid change of the measured property over
time, there is a dynamic error. Often, this behaviour is described with a
Bode plot showing sensitivity error and phase shift as function of the
frequency of a periodic input signal.
• If the output signal slowly changes independent of the measured
property, this is defined as drift.
• Long-term drift can indicate a slow degradation of sensor properties
over a long period of time.
• Noise is a random deviation of the signal that varies in time.
• Hysteresis is an error caused when the measured property reverses
direction, but there is some finite change required for the sensor to
respond, creating a history-dependent offset error.
• The sensor may, to some extent, be sensitive to properties other than
the property being measured. For example, most sensors are influenced
by the temperature of their environment, even if that is not what they are
designed to measure.
15
Such deviations can generally be classified as systematic errors or random
errors:
• Systematic errors can sometimes be compensated for by means of
some kind of calibration strategy. (e.g. the non-linearity of the platinum
resistance thermometer mentioned above)
• Noise is a random error that can be reduced by signal processing,
such as filtering, usually at the expense of the dynamic behaviour of the
sensor. This is discussed further later.
Resolution The resolution of a sensor is the smallest change it can detect in the
quantity that it is measuring. The resolution is related to the precision with
which the measurement can be made.
Sensors The list of physical phenomena that can be measured is very long and
includes:
Acoustic, sound, vibration Chemical, Humidity Electric current, electric potential, magnetic, radio Flow, Pressure, force, density, level Ionising radiation, subatomic particles Position, angle, displacement, distance, speed, acceleration Optical, light, imaging Thermal, heat, temperature Proximity, presence
16
Since there are typically several different sensors per phenomenon then
the list of sensors would be even longer. This is why a description of
sensors would require a whole course. For this reason this course will be
limited to using strain gauges to highlight the main problems that have to
considered when designing analogue signal processing circuits.
Strain gauge
Strain gauges were discussed in the P2 lectures as an example of a
sensor. A strain gauge takes advantage of the physical property of
electrical conductance's dependence on the conductor's geometry. When
an electrical conductor is stretched (within the limits of its elasticity such
that it does not break or permanently deform) it will become narrower and
longer, changes that increase its electrical resistance end-to-end.
Conversely, when a conductor is compressed such that it does not buckle,
it will broaden and shorten, changes that decrease its electrical resistance
end-to-end. From the measured electrical resistance of the strain gauge,
the amount of applied stress may be inferred. A typical strain gauge
arranges a long, thin conductive strip in a zig-zag pattern of parallel lines
such that a small amount of stress in the direction of the orientation of the
parallel lines results in a multiplicatively larger strain over the effective
length of the conductor—and hence a multiplicatively larger change in
resistance—than would be observed with a single straight-line conductive
wire.
17
The so-called gauge factor is defined as:
εG
FRR
G/∆
=
where RG is the initial resistance of the strain gauge
∆R is the change in resistance when strain is applied
ε is the applied strain.
εFG GRR =∆ /
Typically a gauge factors are ~2, which means that the fractional change
in resistance is only twice as large as the strain. This means that the
fractional changes in resistance are very small and they can only cause
small changes in any electrical signals.
18
Signal Conditioning
The output from a transducer is generally a continuously varying or
analogue signal. In contrast digital processors store and process signals
sampled at particular times and represented as binary numbers. Any modern instrumentation must therefore include a component (described in the P2 course), known as an Analogue-to-Digital converter (ADC), which converts the analogue input signal into a digital signal that can be read by the digital processor.
The simplest and cheapest possible instrumentation system is one in which
the output from the transducer is connected directly to the input of an ADC.
However, both the transducer and the ADC are standard components that
have not been designed for any particular application. More importantly,
transducers rely upon physical processes that rarely, if ever, generate
output signals that are compatible with the ADC input range. In particular
the maximum change in the output signal from a sensor is often smaller than the minimum change in signal that can be detected by the ADC. This means that in this simplest system even the maximum
change in the transducer output may be undetectable.
Instrumentation systems must therefore include a circuit before the ADC that amplifies the output from the transducer to make it detectable by the ADC. Such a circuit is referred to as signal conditioning.
19
In instrumentation, signal conditioning generally means manipulating an
analogue signal (from a sensor) in such a way that it meets the
requirements of the next stage of a system for further processing. In
general the most common “next stage” will involve analogue-to-digital
converters.
Signal inputs accepted by signal conditioning circuits include DC voltage
and current, AC voltage and current (and possibly but rarely electric
charge). The processes that are performed by these circuits will almost
certainly include amplification and filtering. In addition to these functions the
signal conditioning may also include a step to isolate the input circuits from
the rest of the system (one area where this is important is in medical
electronics where isolation protects the patient who is hosting the sensors)
and/or a non-linear step (such as a log or an anti-log amplifier) to
compensate for any non-linearity of the sensor.
Although isolation and non-linear stages are key to a few systems
this course will focus on the amplification and filtering that must be
included in almost all systems. Commonly used amplifiers for signal
conditioning.
20
Simple Amplifier Circuits for Signal Conditioning
The output from many transducers is a voltage and there are two simple
circuits (described in the P2 course), the non-inverting and the inverting
amplifier, which can be used to amplify a voltage signal.
Non-inverting amplifier (P2 Revision):
A non-inverting amplifier circuit.
Don’t forget the analysis of any op-amp circuit to understand its function is
based on the simple ideal op-amp rules (taken from P2):
• An ideal op-amp has an infinite input resistance, an infinite differential gain and an output resistance of zero.
The infinite gain means that provided there is negative feedback:
(be sure you understand WHY)
21
This equation and the characteristics of an ideal op-amp can be used to
show that for the non-inverting amplifier
where the gain is given by the terms within the brackets.
Inverting amplifier (P2 Revision):
An inverting amplifier circuit.
Using the same rules as above, and noting that for the inverting amplifier
V+ is connected to ground (0V) we can show that:
inout VRRV
−=
1
2
22
Mini-Summary
Instrumentation systems are widely used to control and monitor many
different “host” systems.
Any physical variable that is being measured, has to be converted to an
electrical signal, usually an analogue signal, by an input transducer or
sensor. Useful information then has to be extracted from this signal, most
often by a programme in a digital processor. Conversion from an analogue
sensor output to a digital input for the processor is a two-stage process
involving analogue signal processing (conditioning) and an analogue-to-
digital converter.
Each instrumentation system therefore usually consists of four constituent
parts, the sensor, analogue signal processing circuits, an analogue-to-
digital converter and a digital processor.
Sensors rely upon physical processes that allow an electrical signal to be
generated in response to a change in a physical variable. These physical
processes usually result in small output signals. A key part of any analogue
signal processing circuit is therefore a circuit that amplifies the changes in
the output signal from a sensor.
23
INTERFERENCE AND INSTRUMENTATION AMPLIFIERS
Introduction
A lot of this material was also covered in the P2 course, but it is important,
so it will be included here.
Inverting and the non-inverting amplifier circuits share a common problem -
they both amplify the difference between the input voltage signal and the
amplifier ‘ground' connection. Any variations in the ‘ground’ voltage will be
indistinguishable from changes in the sensor output voltage. The resulting
interference can be minimised by carefully designing the analogue signal
processing circuit to avoid shared ground connections and coupling to
electromagnetic radiation. In addition, whenever possible the sensor should
be included within a circuit that produces an output that is the difference
between two voltages, a type of output known as a differential output, with
the largest possible amplitude.
A differential output can be amplified using a simple differential amplifier
circuit. However, variations between the actual and nominal values of the
resistors in this circuit will create a response to changes in the average
(common-mode) input signal. To avoid the problems that this causes the
differential output from a transducer is usually amplified using a three
op-amp instrumentation amplifier.
24
Illustration (see also P2 notes): In the analysis of the inverting op-amp circuit presented above you will
notice that there are three points which are shown as ground. Now, this can
present a problem:
An inverting amplifier circuit (showing ground points)
How can we guarantee that these three “ground points” are in fact all at
identical potential? In fact it was an assumption in the analysis, and if the
assumption is not correct then the analysis changes. For example, if the V+
input is not at zero-volts then the V+=0 assumption must be dropped, and
the output voltage becomes:
( ) +++
++−=+−−= V
RRV
RRVVV
RRV ininout
1
2
1
2
1
2 1
It is clear from this that we would not be able to distinguish between sensor
signal (Vin) and ground-point errors. This is a real problem!
There are at least two origins of such problems. In the P2 course the issue
of careful grounding was discussed, illustrated by the following diagram:
25
Figure (6) Schematic diagrams of a poor earthing scheme, on the left,
and a good earthing scheme, on the right.
In Figure (6) the ground connections of three parts of the system on the left
are connected to one another (“chained” together) within the overall circuit
before a single connection is made to earth. A hidden danger with this simple grounding scheme is that each connection has a small but
finite resistance. These resistances are represented by , and
in Figure (6), which shows that this grounding scheme creates
common resistances to ground. The problem caused by these common resistances is that the current flowing to ground through one circuit can change the ground potential of other circuits. As illustrated for the simple inverting op-amp circuit above, it is then impossible to tell the difference between real signals and such changes in ground potential. Since the shared resistances created by this grounding scheme
are formed by contacts, tracks and wires that have an ideal resistance of
zero, it is understandable to think that this effect is negligible. However, the
26
total resistance that is shared could be a few Ohms. If we approximate this
to say 10 Ohms then it would only require a current of 50μA to flow through
this shared resistance to change the 'ground' potential of the input amplifier
by 0.5mV. If the subsequent signal conditioning circuits have an overall
gain of say 1000 this would cause an output voltage “error” of 0.5V. It is
therefore quite possible for changes in the current drawn by one part of the
circuit to create a fluctuation in the 'ground' potential of another part of the
circuit that will be (incorrectly) interpreted as a significant signal.
In the P2 course it was suggested that this interference can be avoided by using a single-point grounding scheme, often referred to as a star connection or star grounding scheme, shown on the right hand side of figure (6).
However, there is a second problem…
Electromagnetic interference
All of the connecting wires act as aerials which can “receive” any
electromagnetic fields in the environment. (This is especially so if there are
any multiple ground paths, which can form beautiful loop aerials!) This can
then result in significant time-varying currents flowing through the ground
connections that could interfere with the circuit ground. Within small-scale
circuits such effects can generally be avoided at low frequencies. However,
in connecting sensors to signal conditioning circuits such electromagnetic
interference on the cables can be problematic.
27
Bridge Circuits
Careful circuit layout reduces the impact of interference arising from undesirable signals on the earth connection of the input amplifier. However, it is better to use a design that prevents the problem by avoiding the need to rely upon a good ground connection. This can
often be achieved by placing the sensor (such as a strain gauge) in a
bridge arrangement, see figure (7), which generates both a signal voltage
and a reference voltage, giving a differential voltage.
In figure (7), the transducer is subjected to the “influence” to be
measured whilst and are reference resistances subjected to
the same conditions as , except for the influence.
Let be the bridge differential output voltage when the transducer
“senses”, such that the sensor resistance changes from R0 to R0 (1+α),
then in this particular configuration, the currents in the potential dividers on
the left and right are:
so that
hence
28
For small values of α this expression shows that the output voltage is
proportional to α and does not include a large constant term. This
means that none of the limited input range of the ADC is wasted representing a large, constant voltage which contains no information.
Rather with this circuit the entire input range of subsequent circuits can be
used to represent the useful signal.
(However, it should be noted that neither of the outputs are grounded, and
this must be kept in mind in designing the next stage of the system.)
Figure 7: A sensor bridge circuit with a differential output.
29
Cables as a source of interference
The bridge circuit with a sensor avoids interference from signals on the
ground connection of the instrumentation system. However, the output signals are still small and are therefore vulnerable to electromagnetic interference. With a pair of wires connecting the bridge circuit to the instrumentation system it is possible for interference to be caused by fields which couple to the connecting wires. There are particular effects
with a connection formed by two wires.
• The first is that the pair of wires can form a loop aerial that couples to any stray changing magnetic field. This can then generate an emf (i.e. voltage) around the loop.
• The second is that if the two wires run parallel, but, separately they will have different coupling capacitances to other conductors in the local environment, particularly the local mains power leads.
To minimise the impact of both these phenomena the two wires are typically twisted around each other to form a twisted pair. This cheap,
and neat, solution ensures that the cross-sectional area of any loop formed
by the wire is minimal and that both wires have the same coupling
capacitance to any other conductor. A twisted pair is therefore often used
as a cheap but effective method of connecting two separate parts of an
instrumentation system.
30
The full bridge circuit
The use of a twisted pair is effective, and represents a good solution to a
problem. A better solution would be to avoid the problem altogether. It is sometimes possible to minimise the impact of interference by adopting a good general design principle. This is to create the largest possible signal at the earliest opportunity. For example, the output signal from a bridge circuit can sometimes be increased by using the full bridge circuit with four sensors, as shown in Figure (8). In this circuit, the sensors are arranged so that each branch
of the circuit contains a pair of sensors acting in complement (for example
one strain gauge in tension and the other in compression). In this way, the
output signal from the bridge circuit is enhanced by a factor of 4.
Figure 8: A full bridge circuit containing four strain gauges.
31
The differential amplifier The output from the bridge circuit, and many other types of transducer, is
the difference between two voltages. These differential voltages will be
smaller than the input range of a typical ADC. A “signal conditioning” circuit
is therefore required that amplifies a differential voltage, rather than a
voltage relative to a ground point. i.e. given two wires coming from a sensor
(or sensor in a bridge circuit) the voltage levels relative to a ground-point
may be not well defined (due to electromagnetic interference, difficulty of
“defining” the ground voltage, etc.). However, the voltage difference should
be representative of the quantity being sensed.
A simple op-amp circuit that can amplify a differential voltage is shown in
Figure (9). This was analysed in the P2 course.
Figure 9: Single op-amp differential amplifier
32
Assume the op-amp is ideal and considering the connections to the op-amp
inverting input:
1
1
2 Rvv
Rvvo −
=− −−
The second input voltage is connected to the non-inverting input through a
simple potential divider, so for this we can write:
The “normal” ideal op-amp rules, together with negative feedback allow us,
as usual, to equate the two op-amp inputs, writing . Using this to
eliminate v+ and v- in the above equations gives us:
The magnitude of the gain is the same as we saw for the inverting amplifier
earlier, but now it is the difference between the voltages which is
amplified, which is just what we need!
For signal conditioning applications a particularly useful aspect of this
amplifier is that the output is dependent only on this difference and is
independent of their absolute levels. This is important, so it is worth
thinking about why this happens.
Consider the situation when the two inputs are the same, let’s say v1=v2=v.
Now the potential divider rule for the non-inverting input remains the same
as before, and still we need , so that means that
33
vRR
Rvv21
2
+== +−
But, the v- input is “fed” by a potential divider between v1 and vo , and given
that v1 is also v the op-amp conditions can only be satisfied if vo is also
zero. This means that the output depends only on the difference between the inputs.
This type of differential behaviour is important, so we develop “standard”
ways of expressing it. Firstly, we write the inputs in a different way. Rather
than expressing the inputs in terms of the two voltages v1 and v2 we
express the inputs in terms of the average and difference of these voltages.
These are referred to as the “common mode” (cm) and “differential” (diff)
voltages, expressed as:
12
12
2vvv
vvv
diff
cm
−=
+=
REMEMBER: these are just a pair of simultaneous equations, so any pair
of voltages can be expressed in terms of either their individual voltages or
in terms of their common-mode and differential terms. We can re-arrange
between the notations using the above expressions for the common-mode
and differential voltages in terms of the input voltages, or their complement:
2
2
2
1
diffcm
diffcm
vvv
vvv
+=
−=
34
CAUTION: although there is no ambiguity in the definition of the common-
mode term there is a sign ambiguity in the differential term, the sign of
which will depend on the labelling of the two individual voltages – be
careful!
Now, going back to the gain equation for the differential amplifier
determined above:
and substituting for v1 and v2 in terms of the common-mode and
differential voltages now makes it obvious that
diffo vRRv
1
2=
which is independent of the common-mode term and has a gain for the
differential term of R2/R1. This differential gain is normally designated as
Adiff.
More generally the output voltage from a circuit intended for differential
amplification can then be written in the form:
cmcmdiffdiffo vAvAv +=
or
where is the differential gain of the amplifier and is the so-called common mode gain. For the ideal differential amplifier
illustrated above the common mode gain is zero, , so that the
35
output is independent of the common mode signal. The differential gain
is obviously (from the above analysis) just 12 / RRAdiff =
Why the circuit might not be non-ideal The above analysis is an ideal situation, but of course in the real world the
differential amplifier might not be so ideal. There are a number of reasons
for this, but most commonly it is because the two resistors labelled R1 will
not be identical to one another and similarly the two resistors labelled R2
will not be identical to one another. You may recall that you analysed this
issue as a tutorial problem for P2! In this analysis you assumed a small
error in the resistors, so that the circuit became something like:
36
where x is a small error (<<1) and is chosen to represent a “worst case”.
Assuming the normal op-amp rules, and after a bit of algebra, in the P2
tutorial you came up with an equation for the output voltage of the form:
( )( )
( )( ) ( )
( )( ) 1
1
2
212
2
1
2
11
111
111
vxRxR
vxRxR
xRxRxRvo
−+
−
++−−
−+
+=
Now, expanding this for small x (i.e. ignoring terms in x2 or higher order)
this can be written:
( ) 11
2
221
12
1
2
21
21
vxRR
vRRRRx
RRvo
+−
+−
+=
then substituting for v1 and v2 in terms of vcm and vdiff , and again assuming
x is small we have:
cmdiffo vRR
xRRRv
RRv
+
−+=
21
1
1
2
1
2 4
so now there is an extra term in addition to the differential gain found
above. The differential gain remains the same ( 12 / RRAdiff = ), but now
there is also a common mode gain:
+
−=
21
1
1
2 4RR
xRRRAcm
37
(which, as expected, obviously vanishes in the limit of perfectly matched
resistors, i.e. when x=0).
In general we would “like” a differential amplified to have a small common
mode gain in comparison with the differential gain. This allows it to “reject”
common-mode signals and amplify the differences. The ability of a differential amplifier to reject any common mode signal whilst amplifying the differential signal, is usually characterised by the ratio between the differential and common mode gains. It is called the common-mode rejection ratio (CMRR)
This means that the common mode rejection ratio of an ideal circuit is
infinity. Despite non-ideal effects the CMRR of a good differential amplifier
is usually large and it is therefore often quoted in decibels
Input impedance When using the basic differential amplifier illustrated above there is another
problem, even if the circuit is carefully engineered with high tolerance
resistors to maximise the CMRR. The two op-amp inputs are at the same
voltage as one another (due to the negative feedback), so any differential
input voltage “sees” the two resistors labelled as “R1”. The differential input
impedance is therefore 2R1. You will recall from P2 that in general when
connecting a signal source (in our case from a sensor/transducer) to an
amplifier it is good to ensure that the input impedance of the amplifier is
38
large compared with the source output impedance. However, this cannot
be guaranteed here. It would therefore be useful to arrange to INCREASE
the input resistance of the differential amplifier.
As discussed in the P2 course the input impedance of the circuit can be increased by simply inserting an op-amp buffer circuit on each input of the differential amplifier as shown in figure (10). The source
(sensor) output (i.e. the input to the circuit) is then directly connected to one
input of an op-amp that has a very high, if not infinite, input impedance.
Provided the op-amps are ideal then in the arrangement shown in figure 10
we have:
which means that the signals from the sensor are applied to the differential
amplifier as before. Adding op-amp buffers to the standard differential
amplifier therefore simply increases its input impedance.
Figure 10: A differential amplifier with buffered inputs.
39
Instrumentation amplifiers In principle we could add further gain to the input stages (buffers) by using
the circuit for the basic non-inverting op-amp amplifier presented earlier.
However, this is unwise because differences between the nominally
identical resistors in the two non-inverting amplifiers will create different
gains on the two input signals. It turns out we can do a bit better than the
circuit in Figure (10) if we connect the non-inverting inputs through a
common resistor, forming the standard 3 op-amp instrumentation amplifier shown in Figure (11).
Figure 11:The standard 3 op-amp instrumentation amplifier.
40
The analysis of this circuit was presented in the P2 lectures, and is repeated here: To understand the circuit first consider the current flowing
vertically through the resistors connecting to
the middle two equations can be re-arranged to give
and the right hand pair give
then subtracting the first of these two equations from the second gives
So, the differential output from the first pair of op-amps (referred to as the
first stage) is equal to the differential input multiplied by a differential gain
factor, let’s say Adiff_1
1
21_
21RRAdiff +=
Adding the equations for and together (and dividing by two) we also
have:
22'' 2121 vvvv +=
+
i.e. the common mode term in the output from the first stage is identical to
the common mode term in the input to the first stage, so the common-mode
41
gain is unity. Thus the first stage can provide a differential gain and unity common-mode gain, thereby increasing the CMRR as well as the input resistance. Additionally, analysis shows that the CMRR of the first stage is not substantially damaged by small errors in the resistor values. Normally the R2 resistors are part of the package and the single R1 resistor is chosen by the user. The output from this first stage is fed to a differential amplifier circuit which is effectively identical to that considered above. This last stage is often given a fixed low differential gain (unity, for example) by the
manufacturer. The advantage of this is that all the resistors within this part
of the circuit can be integrated within the instrumentation amplifier package.
Matching to the required accuracy is then achieved as part of the
manufacturing process. With a low differential gain the main function of the
differential amplifier is to provide a single output whilst rejecting any
common-mode input signal.
The overall CMRR of the instrumentation amplifier can be written in a
number of ways. Given the way that we have expressed the differential and
common-mode gains we can now gather things together. For the first
stage:
( )
+−=−
1
21212
21''RRvvvv
42
and
22'' 2121 vvvv +=
+
or
1_' diffdiffdiff Avv =
and
cmcm vv ='
For the last stage:
( )
+
+
−+−=
2''4'' 12
43
3
3
412
3
4 vvRR
xRRRvv
RRvo
or
cmlaststagecmdifflaststagediffo vAvAv '' __ +=
Hence:
cmlaststagecmdiffdifflaststagediffo vAvAAv _1__ +=
So the overall differential gain is:
1__ difflaststagediff AA
and the overall common-mode gain is:
laststagecmA _
Generally the CMRR is therefore:
1__
_diff
laststagecm
laststagediff AAA
CMRR ×=
43
If the first stage does also have common mode gain then this generalises
to:
1_
1_
_
_
cm
diff
laststagecm
laststagediff
AA
AA
CMRR ×=
In both cases this can also be expressed as:
laststagefirststage CMRRCMRRCMRR ×=
44
Mini Summary
Amplification is required in the vast majority of systems to match the maximum output signal from the sensor to the maximum input signal of an ADC.
Simple op-amp amplifiers with a single input amplify the difference between their input signal and the local ground voltage. Since this
might be different from the ground voltage at the sensor this type of
amplifier is vulnerable to interference caused by fluctuations in the ground potential. When signals are small careful design is therefore
required to prevent interference from sources, including electromagnetic
radiation and current flowing through connections shared by different
circuits. Techniques including shielding, good grounding and twisted pairs can significantly reduce interference.
One approach to creating a system that is robust to interference is to use a
sensitive transducer to create a large signal as early as possible. One
problem that may arise with sensitive transducers is that they can be
sensitive to more than one physical effect, for example a strain gauge may
be sensitive to temperature. In some situations it is possible to design a circuit that distinguishes between changes in two physical quantities that both affect the output of a sensor. In other situations it may become necessary to measure both quantities and then to use the digital processor to correct for the undesirable dependence.
45
Another approach to creating a system that is robust to interference, that
has the added advantage of creating a signal that only represents the
change induced by the physical effect of interest, is to use a sensor with a differential output. This type of sensor must then be connected to a differential amplifier. A simple differential amplifier can be created from a
single op-amp and two pairs of identical resistors. However, variations
between the resistances of nominally identical components leads to circuits
which respond to the average input (common mode) signal.
The problems caused by a finite common-mode gain are usually reduced by using a three op-amp instrumentation amplifier. This circuit has a high input impedance, a high common-mode rejection ratio and an easily set differential gain. An additional attractive feature is
that it is very easy to use!
46
REAL OP-AMPS Introduction (– for information only) The inside of a real op-amp (for example, the 741) is rather
complicated:
This makes the full analysis of a signal conditioning circuit built using
op-amps potentially rather difficult, as the op-amp may not be as “ideal” as
initially assumed!
47
There are two stages in analysing any real op-amp circuit. Initially, the function of the circuit can be determined assuming that the op-amp is ideal. Subsequently, it may be necessary to perform a more detailed analysis of any circuit, including the non-ideal behaviour of the op-amp, in order to quantify the performance of the circuit more precisely. Alternatively, this type of detailed analysis is needed to ensure that an op-amp is chosen which appears to be ideal in the context of a particular circuit.
Behaviour of Real Op-amps
For a real op-amp the gain and input impedance are large and the output impedance is small, however, they are all finite. A model of the op-amp that includes these effects is shown in figure (44). This model of the op-amp can be included in the analysis of a particular circuit to ensure that the non-ideal behaviour of the op-amp selected for a particular application has a negligible effect on the circuit performance.
Figure 44: A model for an op-amp that includes its finite input
impedance, output impedance and gain.
48
In addition, to the finite gain and impedances of the op-amp there are a few
other non-ideal aspects of the behaviour of the op-amp which must be taken
into account when either designing/analysing a circuit or selecting an op-
amp.
Limited Output Voltage Range –Power is supplied to the op-amp via two
power supply connections, not shown on most circuit diagrams. The output voltage of the op-amp is limited by the supply voltages applied to these pins. Thus if an op-amp has a positive supply voltage of +5V and a negative
supply voltage of -5V the output voltage will be limited to the range between
+5V and -5V. In fact the actual output voltage range is likely to be less than
the supply voltage range. If the input conditions require an output voltage
outside the allowed range then the output voltage will saturate to a maximum
positive or minimum negative value. If this range is too small for a particular
circuit then an alternative op-amp should be used which has a larger
maximum supply voltage and hence output voltage range.
Input Offset Voltages – The two input signals are connected to two different
input transistors within the op-amp. Ideally, these two transistors are
identical. However, variations in the manufacturing process mean that this
ideal condition is rarely achievable, which means that for a real op-amp a small differential input voltage is required to create an output of 0 V.
This input offset voltage can be accommodated by one of two alternative techniques. In some op-amps extra connections are provided to allow the user to add a variable resistance that can then be used by the user to zero the input offset voltage. This trimming is the
least expensive technique of compensating for the offset voltage, however, it
49
is inconvenient. For example the resistance may need to be adjusted if the
operating temperature changes. The more usual solution to this problem is therefore for the op-amp manufacturer to include an equivalent resistance within the op-amp package that is adjusted as part of the manufacturing process. The additional manufacturing processes required
to trim a circuit means that these components are more expensive. However,
they are more ideal when received by the user and they are therefore very
popular.
Common Mode Rejection (In the material above the issue of differential and common mode
gains for an op-amp based circuit were considered. Here the situation
for the op-amp itself which is considered.)
The two input signals to the op-amp can be considered to consist of
an average, or common-mode, component and a differential component
so that
The ideal op-amp will only respond to the differential part of these two input signals. However, a real op-amp will respond to the common-
mode signal with a common-mode gain . By convention this aspect of
the performance of an op-amp, and other circuits, is characterised by the
50
logarithm of the ratio of the differential gain to the common mode gain
. This common-mode rejection ratio CMRR is
Many important op-amp circuits are based upon a constant bias, usually 0V,
applied to and a feedback loop connected to . The operation of these
circuits is based upon a high differential gain this ensures that , and
since is a constant this means that is constant. For these circuits the
effect of a finite common-mode gain is therefore negligible. However, in
instrumentation amplifiers the input voltages to the op-amps will vary with the
common-mode signal. For these circuits the CMRR of the op-amp can have
a significant impact on performance.
Input Bias Current - Each input to some op-amps is connected to
the base of one of a pair of bipolar transistors. The base current for
each device flows through the op-amp inputs. These two dc currents
are referred to as the input bias currents. A technique to
compensate for the presence of these currents will be described later.
Noise - Finally, op-amps contain several devices, each of which will
generate noise. This aspect of the behaviour of real op-amps will be
described later.
51
Real Op-amps in an Inverting Amplifier
In order to understand how the finite gain and impedances of an op-amp can
be included in the analysis of a circuit we consider an example, the inverting
amplifier shown in figure (45).
Some definitions before starting the analysis:
open-loop voltage gain (measured without feedback)
closed-loop voltage gain (gain with feedback)
the feedback fraction
Figure 45: Inverting amplifier circuit including non-ideal op--amp model.
52
The voltage at the inverting input to the op-amp can be determined by linear superposition of the voltages created by the two voltage sources within the circuit. To simplify the analysis assume that the op-amp is
almost ideal so that the input impedance is large and the output impedance
is small. In particular, for this circuit assume that and that
.
First to determine the effect of the op-amp output voltage, assume that the input voltage is zero. Then the voltage at the inverting input arising
from the output of the op-amp is equal to vf and is given by:
Then to determine the effect of the input voltage assume that
and that for simplicity also assume and .
Linear superposition means that the voltage at the inverting input can then
be determined by adding the effects of the input voltage and the op-amp
output voltage
++
+−=−= −+
oind vRR
RvRR
Rvvv21
1
21
20
This expression for can now be included in the equation for the voltage
at the output from the op-amp. Taking into account that the input is
connected to the inverting input to the op-amp and the direction of the output
current
53
oodOLo ZivAv −=
and then re-arrangement (eliminating vd ) gives
The closed-loop gain is therefore given by:
but since
this expression becomes
Finally, if then
The conclusion is that:
(i) the output impedance has decreased by a factor of
(ii) as long as the op-amps open loop gain is greater than the closed
loop gain the closed-loop gain is independent of the exact value of
the op-amp's open-loop gain.
54
Real op-amps in a non-inverting amplifier
Figure 46:A non-inverting amplifier circuit including
non-ideal op-amp model.
The circuit shown in figure (46) can be analysed to include the effects of a
non-ideal op-amp, see the tutorial problem, to show that;
This means that the circuit behaves as an amplifier with closed-loop gain:
and the output impedance is
The expression for the closed-loop gain can also be re-written as follows:
55
If , then and the closed-loop gain once again only
depends upon the components in the feedback circuits and is independent of
the exact value of as required.
Since , the condition is equivalent to .
Thus one of the conditions for selecting an op-amp is that the open-loop gain
is much larger than the closed-loop gain of the final circuit.
Input impedance- Assume that the circuit design is power efficient,
so that any currents through the feedback resistors are small
compared to the current delivered to the load impedance ZL, then Zo
and ZL form a potential divider and the output voltage becomes:
oL
LdOLo ZZ
ZvAv+
= (a)
Now the voltage v2 can be determined from v0 and the R1 and R2
potential divider, so:
021
102 v
RRRvv β=+
=
where β is R1/(R1+R2). Substituting in for vo from eq. (a) we have:
doL
LOL v
ZZZAv+
= β2 (b)
Now, considering figure 46, the input voltage vin can be written as the
sum of v2 and vd , i.e.:
din vvv += 2
Substituting in for v2 from eq. (b) we then have:
56
+
+=++
=oL
LOLddd
oL
LOLin ZZ
ZAvvvZZ
ZAv ββ 1
We can now determine ZinCL by dividing by the input current iin
+
+==oL
LOL
in
d
in
ininCL ZZ
ZAiv
ivZ β1
Thus
Finally, assuming that , so that most of the power from the op-amp
output is delivered to the load and not dissipated internally.
so the already large input impedance is increased by the feedback loop.
To summarize:
(i) The output impedance has been decreased by a factor
(ii) The closed-loop gain equal to which is independent of the
exact value of the op-amp's open-loop gain.
(iii) The input impedance has been increased by the factor
and the conditions for selecting an op-amp to ensure that an op-amp
appears to be ideal for this application are
57
Correcting for finite input bias currents
When designing an op-amp circuit it is often necessary to consider the
finite input bias current which many op-amps require. The input impedance
of the op-amp that has been considered so far represents the changes in
this current that occur when the two input voltages change. It is also
necessary to consider the average current that must be supplied to these two inputs. Each of the inputs pins may require a dc bias current to
ensure that a bipolar transistor within the op-amp operates correctly. These bias currents can be represented by two current sources between the relevant input and ground, shown in figure (47).
Figure 47: An inverting amplifier circuit including input bias currents. For an
ideal op-amp only two resistors are required, however, a third resistor is
required to compensate for the input bias currents.
58
In the example circuit shown, an inverting amplifier, there are three signal
sources, , and that determine the voltage at the inverting input
to the op-amp. Superposition means that the total voltage at the inverting
input, can be calculated by adding together the effect of each source. During
the calculation of the effect of each signal source, any other voltage source
is short-circuited and any other current source is assumed to be open circuit.
This means that whilst calculating the effects of the two voltage sources, the
bias current source is open circuit, i.e. the op-amp is ideal. The error in
voltage at the inverting input caused by the presence of is therefore the voltage generated at this circuit node when the two voltage sources are short-circuited, i.e. both voltage sources are set to zero. With the
output voltages from both the voltage sources set to zero then both and
connect the inverting input to ground. The bias current flowing through
this parallel combination of resistors will then cause a voltage signal
At first it would appear that an input bias of an op-amp would be
insignificant. However, this expression shows that with typical resistance
values of 10K or more, an input bias current of 1μA can generate an error
of 10mV. This is significant compared to the signal from the typical sensor.
There are two possible solutions to this problem:
(i) select an op-amp with a small bias current or
(ii) use small value resistances in the feedback circuit.
59
Unfortunately in some situations neither of these two solutions are possible. In this case the circuit can be modified to significantly reduce the impact of the undesirable signal.
To understand how the circuit must be modified remember that the
op-amp has a high, if not quite infinite, differential gain. It will therefore
amplify any differential input voltage, such as the one created by the bias
current flowing in resistors and , even in the absence of an input
signal. The effects of this bias current will become negligible if the inverting
input is at the same voltage as the non-inverting input so that the differential
voltage is zero. Then there will be no differential input voltage when the input
to the circuit is ground. The effects of the input bias currents will then be
negligible.
In the ideal circuit the voltage at the non-inverting input is zero, even when
there is a finite input bias current. To allow the input bias current to create a finite input voltage a resistor is required between the non-inverting input and ground, see the dashed box in figure (47). Assuming
that the resistance of this extra resistor is , then . However,
in order to obtain an output voltage of zero when the input is zero the value
of should be chosen so that under these conditions . Thus
and since in op-amps the two inputs to the op-amp will be connected to
nominally ‘identical’ devices we can assume that and hence
60
that is the new resistor is equivalent to the other two resistors acting in
parallel. The addition of this extra resistor, which would have no effect on an
ideal op-amp, is therefore to cancel the effects of the finite input bias current.
Frequency Response
So far it has been assumed that the gain of an op-amp is independent of
frequency. However, because of unavoidable parasitic capacitances that
form part of each transistor within the op-amp the gain of a real op-amp is
frequency dependent. The result is a complex frequency dependent gain
(and phase shift) that is both difficult for the manufacturer to control and/or
specify accurately and for the designer to use confidently.
The solution to this problem that has been adopted by manufacturers is to
include a large capacitance within each op-amp. This capacitance is then
designed to ensure that over the frequency range of interest the op-amp
behaves as if it had a single dominant RC response. Thus compensated op-amps (such as the 741 and all the op-amps that you are likely to use) have the frequency response characteristics of a low-pass RC filter
where is the dc gain of the op-amp and is the 3dB break-point
frequency created by the frequency compensating capacitor.
61
The op-amp is then designed to ensure that its gain falls to unity at a frequency that is low enough so that any other frequency dependant responses within the op-amp are negligible. The consequence of this
is that the 3dB break-point frequency for compensated op-amps is extremely low: 5 Hz for the 741, for example. The convention in an op-amp data sheet is to give the values of the dc
gain, , and the unity gain frequency (which is the frequency at which the open-loop gain has fallen to 0 dB). This frequency
represents the maximum frequency at which the op-amp could be used to
create a unity gain buffer. It is therefore a measure of the maximum useful
frequency of the op-amp.
The closed-loop frequency response of any circuit can now be
determined by replacing the frequency independent open-loop gain by
the frequency dependant gain. Thus for the non-inverting amplifier
becomes
Because for this particular circuit the feedback network only contains
resistors is independent of frequency and is a straight line parallel
to the frequency axis, as shown in figure (48).
62
Figure 48:Closed-loop frequency response of a non-inverting amplifier.
At the point Y in figure (48) and hence at this point:
The bandwidth of the circuit with the feedback loop, the closed-loop
bandwidth, is therefore given by:
63
Thus although the open-loop bandwidth of the op-amp alone is only , the
bandwidth for the circuit with feedback is where, assuming ,
The greater the amount of feedback (i.e. the greater the value of and the
lower the gain ), the greater the bandwidth . In fact this equation
shows that the gain bandwidth product of the closed loop circuit
is a constant and the value of the constant is determined by the op-amp.
64
Frequency compensation and slew rate
One effect of the inclusion of a compensating capacitor within the op-amp is that it limits the rate at which the op-amp can respond to any sudden changes of input. In the data sheets the parameter used to
characterise this effect is the slew rate (usually expressed in ) which is the maximum rate of change of the output voltage of the op-amp.
Op-amp data sheets usually give the slew rate for unity gain. Assume,
therefore, that the input to a voltage follower is a large-amplitude high-
frequency sinewave where:
The maximum rate of change of the output is given by:
For an output free of distortion, the slew rate determines the maximum
frequency of operation for a desired output swing. Thus:
where is the slew rate in V/μs for a unity gain circuit.
65
Mini-Summary
An ideal op-amp has an infinite differential gain, an infinite input impedance and an output impedance of zero. The characteristics of an ideal op-amp are used to select a circuit which performs the required function. Real op-amps have a large differential gain, a large input impedance and a small output impedance. These are represented in a model of the op-amp that can be used to derive conditions for an op-amp to appear to be ideal in a particular circuit. Other aspects of the behaviour of real op-amps also affect the performance of circuits. In particular the output voltage is limited by the voltages used to power the op-amp. There is also a finite dc input bias current whose effects can be cancelled by including an additional resistance in the circuit. The gain of a real op-amp is frequency dependant. To simplify the design of circuits the op-amps are designed to have a single dominant pole and therefore the gain-bandwidth product of an amplifier including an op-amp is constant. The slew-rate of the op-amp limits the maximum rate of change of the output voltage of an op-amp.
66
NOISE
Introduction
So far the emphasis has been on how to amplify an analogue signal so that it matches the maximum input voltage of an ADC. This will determine the maximum signal that can be detected without saturation. It is at least as important to know the minimum reliably detectable signal.
Usually, the minimum detectable input signal is approximated to the
maximum error that is made during the conversion process. This is half of
the change in input voltage that corresponds to a change in output of one
least significant bit (LSB). For a 12-bit converter with a maximum input
voltage of 5V this error is
.
In itself this is a small voltage that shows that the system is vulnerable to interference. However, the original source of this ADC input signal is a sensor output signal that is considerably smaller. Thus for example if the maximum transducer signal was itself 10 mV, i.e. an amplifier with a gain of 500 has been used between the transducer and the ADC, then 1/2 LSB is equivalent to a change in transducer output of 1.2 μV. Careful design is then required to ensure that this small signal can be reliably detected despite interference and other signals generated by various unavoidable physical processes within components. Assuming that interference
67
can be avoided it is the noise signals, caused by the physical processes underlying charge flow in components, that determine the sensitivity of modern instrumentation systems.
Fundamental noise sources
There are three noise mechanisms that are found in many components,
Thermal or Johnson noise
Figure 24: White noise from a 100 kΩ resistor in a bandwidth of 10 kHz.
Any resistor generates a noise voltage across its terminals, known as
thermal or Johnson noise. This type of noise arises because the
measured macroscopic current actually represents a flow of charge
carriers through a component. The average velocity of these carriers,
68
and hence the average current, is determined by the electric field, however, interactions between carriers mean that there are variations in the velocity of individual carriers. These fluctuations in the motion of charge carriers in the resistor then lead to fluctuations in the current flowing through the resistor, which translate into momentary random changes in the voltage across the resistor such as those shown in figure (24). In the frequency domain these random
fluctuations correspond to equal amounts of power at all frequencies and
it is this flat frequency spectrum which leads to this type of noise also
being known as white noise.
The flat frequency power spectrum means that the total amount of noise power will be proportional to the noise bandwidth of the
system B . It is therefore not surprising that the root-mean-square output
voltage generated by a resistor is proportional to B . In fact for a
resistance R
where k is Boltzmann's constant and T the absolute temperature.
To estimate the voltage change that this can represent, consider a 100 kΩ
resistor at room temperature, over a modest bandwidth of 10 kHz:
This is a small voltage, however, it is not negligible compared to the signal voltages that may need to be reliably detected in any application which requires a sensitive measurement. The impact of
69
this type of noise should therefore be considered in any carefully designed
instrumentation system.
The equation for thermal noise shows that two general principles should
be followed in order to reduce the amount of Johnson noise.
(i) Any resistances should be kept as small as possible.
However, the ability of the designer to follow this strategy will always
be limited by the necessity to limit power consumption and to ensure
that any parasitic resistances in the circuit, such as contact and
track resistances, are negligible. This usually limits the smallest
resistance used to be greater than 1 kΩ.
(ii) The frequency bandwidth of the system should be limited to the frequency range of interest. Filters should therefore be used
to limit the frequency range of the ADC input signal to the frequency
range required to capture any necessary information. Thus for
example if vibrations in the range 0.1-1kHz from a motor are
indicative of wear which requires maintenance then the system
should be designed to limit the signal to frequencies in the range
0.1-1kHz. The filter circuits used for this will be discussed later.
In addition to these general precautions it is sometimes necessary to calculate the effects of Johnson noise in a particular circuit. The equivalent circuit for Johnson noise that is used in these calculations is an ideal noise-free resistor in series with a voltage
70
noise source or in parallel with a current noise source as shown in figure (25).
Figure 25: The Thevenin and Norton equivalent circuits to represent noise
in a resistor.
Shot noise
Like thermal noise, shot noise also arises because the measured current
represents a flow of charge carriers through a device. Thermal noise arises from instantaneous variations in the average velocity of carriers. Shot noise arises from fluctuations in the number of carriers. The result of these microscopic fluctuations in carrier flow can be
represented by a macroscopic current noise source . If a current is
flowing then for a noise bandwidth
where
Shot noise occurs in diodes, but, most importantly it arises as a
consequence of the small, but, finite base current of a bipolar transistor.
71
This is an important contribution to the noise in some op-amps that will be
considered later.
1/f or Flicker noise.
Flicker noise is present in many different physical systems. In electronic
components it is associated with fluctuations in current flow caused by
temporary trapping of charge carriers.
Unlike the other two noise sources that have been mentioned the flicker
noise power per unit frequency depends upon frequency. Thus when a dc current I is flowing the noise in a frequency range of 1Hz centred on frequency f, is
K and α are device dependant constants.
To calculate the total noise between two frequencies and the
expression for is integrated to give
this shows that each decade of frequency adds an equal amount to the total noise of the device.
The most important aspect of 1/f noise is that it shows that the
significance of 1/f noise decreases as the frequency of interest increases.
In all devices it will therefore always be less than the thermal noise at all
72
frequencies above a critical frequency. The value of this critical frequency
depends upon the particular device. However, there are some important general trends.
(i) 1/f noise is most significant in MOSFET transistors. These
form the basis of almost all digital circuits and are increasing used to
design analogue circuits on the same substrate as digital circuits.
For this type of device 1/f noise can be the dominant noise source
for frequencies up to 10kHz or above. 1/f noise therefore usually has
to be taken into account in noise calculations.
(ii) Compared to thermal noise, 1/f noise is only important in most bipolar devices at frequencies below 10-100Hz. Its effects
are therefore usually insignificant, particularly when low-frequencies
are removed by filtering.
(iii) 1/f noise is usually negligible in resistors, however, normal carbon composite resistors can have an order of magnitude more 1/f noise than a wire-wound resistor. In critical parts of a
circuit it is therefore sometimes necessary to use expensive wire-
wound resistors rather than the more usual carbon composite
resistors.
73
Noise in Ideal Capacitors and Inductors An ideal inductor has no resistance and therefore it will not generate noise. Noise arises from fluctuations in either the average velocity or
number of carriers. An ideal capacitor contains a perfect insulator that blocks the flow of carriers, which means that an ideal capacitor will not generate any noise. Addition of noise sources
Noise arises from random processes within devices, which cause
voltages and currents to fluctuate around mean values. Since it is impossible to ascribe a meaningful frequency or phase to this type of random signal, the only parameter that can be used to characterise noise is its power or root mean square amplitude.
The quality of a signal in the presence of noise is most often specified by the signal-to-noise ratio (SNR)
where is the rms value of the signal and is the rms value of the
noise.
A critical step in determining the SNR at the output of a circuit is to
calculate the total rms noise arising from the different noise sources within
the circuit.
74
For two noise sources the total mean square output voltage is
which can expanded to give
If the two noise sources are independent, they will be uncorrelated and
so that
Since the fluctuations in different devices will be independent this equation can be used to add the noise from different devices. The
first important consequence of this can be highlighted by a simple
example. Consider a situation in which the rms output voltages of two
noise sources are and , then the total rms output will be
.
This clearly demonstrates that if there are two unequal noise sources then the larger noise source will dominate. In order to reduce the amount of noise in this situation the designer should concentrate upon reducing the noise from the larger noise source. A good strategy when designing a system is therefore to ensure that all noise sources are approximately equal.
The equation
75
can also be used to add the effect of two noise sources in the same circuit. However, before it can be used the effect of each noise source within the circuit must be calculated. This can be done by simply assuming that each source acts alone in a circuit in which all other voltage sources are short-circuited and other current sources are open-circuited. This first stage of these calculations is therefore similar to using superposition to calculate the response of a circuit to several signal sources. This principle states that in a linear circuit the response of two or more sources acting simultaneously is the sum of the responses for each source acting alone with the other voltage sources short-circuited and other current sources open-circuited. However, in the principle of superposition it is implicitly assumed that there is a fixed relationship between the phase of the various sources. The overall response is therefore calculated by simply adding the individual responses. This assumption is not valid for the signal sources representing noise that contains many frequency components. In this case the equation
must be used to calculate the total amount of noise.
76
Figure 26:Two resistors in parallel, including the current sources that
represent noise generated within each resistor.
An important simple example of a circuit containing two noise sources is a
parallel combination of two resistors, and . First calculate the noise
voltage caused by noise source, . For this calculation is assumed
to be open-circuit, i.e. disconnected or removed from the circuit. The
current therefore flows through the parallel combination of and .
The corresponding output voltage is therefore
and similarly
The total noise can then be calculated
However, for a resistance R
77
which means that
This is the noise voltage arising from a resistor with the same resistance
as the parallel combination. In this example it would therefore be easier to
combine the resistors to determine the effective total resistance and then
calculate the corresponding noise. In fact this is an example of a general
principle. A simple approach to calculating noise in networks of resistors is to calculate the effective resistance and then add a single noise source to represent the noise of the whole resistor network.
78
A circuit model for op-amp noise
Figure 27: Circuit model to represent noise generated within an op-amp
Calculation of noise in a circuit that includes an op-amp requires a model
for noise generated within the op-amp. A noisy resistance can be
represented by an ideal resistance and a voltage or current source that
generates the noise. Similarly, the noise produced internally by an op-amp is modelled, by a noiseless op-amp and three noise generators. As shown in figure (27) these are a noise voltage generator and two noise current generators.
79
Figure 28 Circuit diagram showing the various noise sources
in an inverting amplifier circuit.
This model can be used within any circuit to calculate the total noise. For
example consider the inverting amplifier circuit, which includes a resistor
connecting the non-inverting input to ground to compensate for the input
current. Figure (28) then shows the circuit including the model of a noisy
op-amp.
The first stage in calculating the total equivalent noise at the input is to calculate the effect of each of the individual signal sources.
Start by calculating the noise at the non-inverting input. Both the
noise current sources, and , create a noise voltage when
they flow through resistor . Adding these two voltages to the
80
noise voltage from the op-amp itself leads to an expression for the total voltage noise at the non-inverting input
Now in the circuit attached to the inverting input there are two resistors. These can be combined into an equivalent circuit before
proceeding. Since the voltage source at the output of the op-amp is assumed to be short-circuited during the calculation of the signal
from all the noise sources the resistors and appear in parallel to create an effective resistance
This effective resistance both generates its own thermal noise and
acts as the effective resistance to ground for the current noise .
The noise voltage at the non-inverting input is therefore
The final step of the calculation is to combine the noise at the two input pins to give the total noise at the non-inverting amplifier input
Assuming that this reduces to
81
To simplify this expression further assume that the amplifier circuit has
been well designed so that 1
. Hence
this can be further simplified using the expression
This last expression shows the three contributions to the total noise at the input to the amplifier; (i) the voltage noise generated by the amplifier
(ii) the voltage noise generated by the current noise in the
amplifier flowing through an effective resistance, , which
arises from a combination of the resistors associated with the
amplifier
(iii) the thermal noise generated by the effective resistance of the amplifier circuit
The importance of the existence of the current noise, which primarily
arises from shot noise associated with the input bias current, is now clear.
If is small then the total noise will be dominated by the voltage
noise of the amplifier. However, as increases the contribution
from the other two terms will also increase and in particular the
1The resistance R3 has been included to compensate for the DC bias
current that flows into the inputs of some non-ideal op-amps (see
earlier section.)
82
contribution from the current noise of the amplifier, which is proportional to
, will begin to dominate. This result demonstrates the
importance of minimising . Unfortunately, the minimum value of
will be restricted by the output resistance of the source of the input
signal, . This arises because the values of R1 and R2 in the circuit will
be designed to be large compared to . This is necessary to both
reduce signal loss arising from the current drawn through and to
ensure that the amplifier gain is determined by R1 and R2. For an
amplifier, for which R2>R1 these requirements mean that R2>R1>>RS.
Then, since R3 is equivalent to the parallel combination of R1 and R2,
and there is a limit on the minimum value of . This is one reason why it is important to select a sensor with a small output resistance. If this is not possible it may be necessary to select an op-amp that has a small input bias current and therefore a small noise current.
The noise at the input to the amplifier circuit is indistinguishable from the input signal. Like the signal it will be amplified by the gain of the amplifier circuit. Unfortunately the presence of noise in both the
op-amp and the resistors in the feedback circuit reduces the signal to
noise ratio at the output of the amplifier compared to that at the signal
source. Since the SNR will only be reduced by any analogue circuits it is important to use a sensitive sensor so that the input SNR is as large as possible. This is another advantage of using a full-bridge circuit
that can amplify the signal without adding any extra noise.
83
Quantisation Noise A further mechanism that degrades the signal-to-noise ratio in an instrumentation system is the errors introduced by quantisation in the analogue-to-digital converter.
In an n-bit ADC every analogue input voltage is represented by a digital
output that corresponds to one of analogue reference levels. The maximum error that will be made during each conversion is half the difference between two of these levels. If the full-scale input voltage of
the ADC is , then the difference between two reference levels is
and the corresponding maximum error is . With the ever improving performance, and reducing cost, of ADCs it is now quite common for systems to be designed so that this error is smaller than the input noise. The output from the ADC will then be accurate enough to represent the noise as well as the signal.
For the situations in which cost means that it is not possible to use an
ADC with a large number of bits, it is possible to calculate the variance of
the error introduced by the quantisation process. This error is often
referred to as quantisation noise. It can be treated as another voltage noise source acting at the input to the ADC, however, the effects of this can only be quantified using an expression for the variance of this error.
84
To calculate this variance it is conceptually easiest to consider a flash
converter, however, the result will apply to any converter. Let the voltage
difference between reference levels in the ADC be δ. Then the error, e, for
each conversion will be between δ/2 and -δ/2. Assuming that each error is
equally probable, then the probability density of error e, p(e), is
Since the mean value of the error is zero, its variance is
and hence
This suggests that the ADC is more accurate than the simple estimate based upon the largest error, δ/2, suggests. However, the
difference is relatively small. In many situations it is therefore appropriate
to use the estimated error δ/2, but, to remember that this is a conservative
figure so that if an n-bit ADC appears to be just adequate based upon the
simple estimate, it will be adequate because the quantisation noise has
been overestimates by a factor of 1.7, which is almost the equivalent of an
extra bit.
85
Mini Summary Noise signals arise from unavoidable fluctuations in the flow of charge carriers through a conductor. The small signals that this
generates can be comparable with the signals that can be represented at
the output of modern ADCs. Noise signals can therefore be critical in determining the accuracy of modern instrumentation systems that are no longer limited by ADC performance.
Thermal noise arises from fluctuations in the average velocity of carriers in a conductor. This is the dominant unavoidable noise source
in any resistor in a circuit. It can only be limited by either reducing the
value of resistance used or limiting the bandwidth of the signal to the
frequency range required to capture any necessary information.
Shot noise arises from fluctuations in the number of carriers passing a particular point in a device This mechanism is important in bipolar
devices and hence in any op-amp which incorporates these transistors. As
with thermal noise shot noise can be reduced by limiting the signal
bandwidth.
The power per unit bandwidth of flicker (1/f) noise reduces as frequency
increases. In most devices this noise source is only important at low
frequencies. However, it is important in MOSFET devices that are
increasingly used to design analogue circuits in the same package as
digital circuits.
86
Noise is a randomly fluctuating signal which is characterised by its root
means square amplitude.
Ideal capacitors and inductors do not generate noise themselves.
The quality of a signal in the presence of noise is most often specified by
the signal-to-noise ratio (SNR) which is
where is the rms value of the signal and is the rms value of the
noise.
If the two noise sources are independent, they will be uncorrelated
which means that
and the root mean square output of two noise sources is
This means that a good strategy when designing a system is to ensure
that
all noise sources are approximately equal.
A simple approach to calculating noise in networks of resistors is to
calculate the effective resistance and then add a single noise source to
represent the noise of the whole resistor network.
If a circuit with a large output resistance is connected to the input of an
87
op-amp amplifier the total noise of the circuit can be dominated by the
current noise of the op-amp. It is therefore important to minimise output
resistances. When this is not possible it may be necessary to select a op--
amp with a zero input bias current, and hence a negligible current noise.
The standard deviation of quantisation noise, which represents fluctuating
errors caused by the process of digitising a signal, is
In order to achieve a requirement for both high-gain and low-noise it is
often necessary to have a low-gain, low-noise circuit, usually called a pre-
amplifier, followed by a high-gain circuit.
88
NON-IDEAL DIGITAL-TO-ANALOGUE
AND ANALOGUE-TO-DIGITAL CONVERSION
Introduction Specification of D/A converters
Figure 12: The ideal response of a 3-bit DAC, showing the analogue
output voltage as a fraction of the full scale output FS. Each bar
represents the output for a particular input and the dashed line shows
the line connecting the ideal outputs.
As you learnt in the P2 course last year, a digital to analogue converter (DAC) converters a digital input represented as a binary number to an analogue voltage (or current) that is proportional to the
89
value of this input. The ideal relationship between the analogue output
and digital input for a 3-bit converter is shown in figure (12). 2
Various physical processes occur when circuits, including DACs, are
manufactured which mean that it is very difficult, if not impossible, to manufacture a circuit which achieves the ideal performance specification. In the case of DACs the resulting error is characterised as the maximum deviation between the actual and ideal outputs. This absolute accuracy is expressed as a fraction of the output change caused by a change in the digital input of one least significant bit (LSB). Offset Error Gain Error
Figure 13: Exaggerated examples of DAC offset error, on the left,
and gain error, on the right.
2 In figure 13 only 3 bits have been shown for clarity. However, in real instrumentation systems
DACs with 6, 8, 10, 12 and 14 bits are often used.
90
The maximum deviation of the actual output from the ideal output is the
absolute accuracy of the DAC. There are several different types of error
that might occur in the output of a DAC. An offset error means that the error between the actual output and the ideal output is the same for all binary inputs. In contrast, a gain error means that the slope of the ideal and actual outputs are different, see figure (13).
Figure 14: The response of a DAC showing the two lines used to define
the non-linear response of the DAC.
Even for a system with no offset error and an ideal gain the individual
outputs may still deviate from their ideal values. These errors are
characterised as the integral non-linearity (INL) and the differential nonlinearity (DNL). The INL is the maximum difference between the actual and ideal output calculated for each digital input. As with all errors the INL should be less than 1 LSB. The DNL is a measure of
91
the changes in output between successive inputs expressed as a fraction of the LSB. Again this should be less than an LSB. However,
this may not be achieved and in an extreme situation the DNL may be
more than –1LSB. In this case for a particular digital input an increase by
1 LSB in the input will cause a decrease in the output. This might be
acceptable, but, it has to be avoided in many applications, particularly
when the DAC output forms part of a control system or an ADC.
Clearly, for an n-bit DAC to be credible all errors should be less than LSB/2. This will almost certainly be correct for low-frequency operation.
However, there are other sources of error that can degrade the DAC
performance at high frequencies. The most common cause of high
frequency errors are glitches. These are spikes in the output which occur
at transitions between different inputs because of imperfections in the
DAC circuitry, for example during a transition from 0111 to 1000 the most
significant bit (the MSB) may change fractionally faster than the other bits
so that there is an instant at which the output corresponds to 1111. These
glitches only become important when the difference between the
switching times of different bits become a significant fraction of the time
for which each different digital input is applied. These glitches are critical
to an increasing number of systems that rely upon a digital circuit and a
DAC to create a well controlled, programmable output signal (An example
of this type of system is the signal generators which you use in labs.).
Despite these applications many data sheets for DACs still specify the
maximum operating frequency as the highest frequency at which the DAC
output attempts to change in response to a new input. This may be
dramatically higher then the frequency at which the absolute error is less
92
than an LSB. All data sheets therefore have to be read carefully when selecting a component to ensure that the required performance will be achieved at the required frequency.
Specification of A/D converters
As with DACs, the ADC characteristics are rarely ideal. Once again the
ADC performance is specified in terms of offset error and gain error
together with integral and differential non-linearities. The offset and gain
error are each defined in terms of the input voltage at which the code
transitions occur. Ideally, as shown in figure (19), for an ADC with a
maximum input of these transitions occur when the ratio is
an odd multiple of ½ LSB.
Figure 19: The ideal response of a 3-bit
ADC.
93
The offset error is the difference between the input corresponding to the first code transition and ideal value. The gain error is the difference between the first and last transitions and the ideal value for this parameter.
Non-linearities of an ADC are defined with respect to the code centre line which is the line joining each of the mid-points of the measured code ranges shown in figure (20). The integral non-linearity is then the maximum difference between the code centre line and its ideal location. Similarly, the differential non-linearity is the maximum difference between neighbouring code transitions.
Figure 20: The response of an ADC showing the differential and integral
non-linearities and a missing code.
94
Finally, figure (20) also shows one last possible error. In some situations it
is possible for the ADC to have a missing output code that is never
generated.
Oversampling Converters
Over the past few years a new type of conversion architecture has
emerged for low and medium speed applications, for example high-quality
digital audio. These new architectures exploit the increasingly low cost
and high clock frequencies available in digital circuits to sample an input
signal at many times the rate required to represent its maximum
frequency. This oversampled digital signal is then processed by
algorithms, known as digital filters, that filter the signal to reduce its
bandwidth whilst increasing the number of bits representing the signal.
The advantages of this approach are:
(i) The requirements placed upon the accuracy of the analogue
components of the converter are reduced.
(ii) The digital filtering after sampling allows the requirements on
the anti-alias filter to be relaxed.
(iii) In many applications a sample-and-hold circuit is no longer required.
95
Mini Summary
DAC's and ADC's are important components of instrumentation and control systems. They convert analogue input signals into a digital format and enable the generation of analogue output signals to control actuators.
Like all electronic components the actual performance of real DACs and ADCs are different from their ideal performance. Several different
measures of these performance errors are used to characterise each type
of DAC or ADC. This information is contained in a component data sheet.
However, these data sheets need to be read carefully when selecting a
particular component to ensure that it will perform to the required
accuracy.
96
FILTER CIRCUITS Introduction A filter is a circuit whose transfer function, that is the ratio of its output to its input, depends upon the frequency of the input signal. The
resulting frequency selectivity of filters means that they are used to fulfil a
variety of functions in instrumentation and signal conditioning systems. In
particular they are a vital part of any well-designed analogue signal
processing circuit between the sensor and the ADC.
Filters were part of the first year P2 course, so much of the material should
be familiar.
Basic filter ideas
There are a number of ways to classify filters, but the simplest way is to
classify them in terms of their frequency-dependent transfer function. We
can write this as:
( ) ( ) ( )ωωω jVjGjV inout =
where “G(jω)” is the transfer function.
Note: Although this is very simple, there are variations in the way this
transfer function is presented. For example, sometimes it is written without
97
the “j” in front of the ω, i.e. ( ) ( ) ( )ωωω inout VGV = . The advantage of including
the “j” is that the same form can then be used with Laplace transforms, i.e.
( ) ( )sGjG →ω
The transfer function contains “information” about how the amplitude and
phase of the signal is influenced when passing through the filter. It is
common to “classify” filters in terms of how the amplitude is influenced. Many
types of filter can then be classified, but the most important are:
Low-pass filters allow any signal at a frequency below a characteristic frequency to pass (ideally unattenuated).
High-pass filters allow signals above a characteristic frequency to pass (ideally unattenuated). Band-pass filters allow frequencies in a particular range to pass (ideally unattenuated). Band-stop filters block frequencies in a particular range (ideally fully attenuated).
These are illustrated in the figure below:
98
Of course, practical filters do not have ideal pass/block properties – therefore
it is important to consider the behaviour of various basic filter elements.
First order filters
The most basic filter elements are first-order filters. This can be either low-
pass or high-pass – other functions are not available with first-order filters.
The basic first order low pass filter consists of a resistor and capacitor:
You should be very familiar with the analysis of this circuit (remind
yourselves if not!), leading to a transfer function between the input and
output of the form:
( )0/1
11
11
1ωωωω
ωjTjRCj
jG+
≡+
≡+
=
where RC is equal to the time constant T (sometimes given the symbol τ)
and the reciprocal of the cross-over (or 3dB) frequency ω0 .
The behaviour of this basic filter element can be represented in various
ways:
99
Here, (a) is a Bode-plot of the frequency response of the amplitude (A) and
phase angle (φ) of the transfer function, where the time-constant is set to be
equal to unity. Logarithmic axes are used for the amplitude and frequency,
showing that above the cross-over frequency the amplitude of the transfer
function drops by a factor of ten for each factor of ten increase in frequency.
This is equivalent to 20dB per decade (remember dB is a logarithmic power
scale). We can determine the magnitude (or amplitude) of the transfer
function as an equation very simply:
( ) 20
20 /1
1/1
1ωωωω
ω+
=+
=j
jG
which at high frequency is equivalent to ω0/ ω , i.e. inversely proportional to
ω, hence the 20dB per decade drop-off.
100
The plots (b), (c) and (d) show the amplitude responses as a function of time
to a step, impulse and ramp input – these will be important when you
consider control systems.
Plot (e) is a polar plot of the transfer function. To produce this plot the real
and imagninary components of the transfer function are plotted on an argand
diagram as the frequency is varied. For the basic first order low-pass filter we
can see that when ω=0 we have G(jω)=1, and as ω tends to infinity G tends
to zero. Another “easy” point to see is that when ω=ω0 we have
2/2/1 jG −= . As ω varies from zero to infinite frequency we follow the
semi-circle from unity around to zero. This polar notation for transfer
functions will also be important when considering control systems.
The basic first-order high-pass filter can also be constructed with a resistor
and capacitor:
Again, you should be very familiar with the analysis of this (remind
yourselves if not!), leading to a transfer function between the input and
output of the form:
101
( )0
0
/1/
11 ωωωω
ωω
ωωω
jj
TjTj
RCjRCjjG
+≡
+≡
+=
where RC is again equal to the time constant T (sometimes given the symbol
τ) and the reciprocal of the cross-over (or 3dB) frequency ω0 . Similar to the
diagrams of the low-pass filter transfer function and responses, we have:
Now (a) shows a 20dB per decade cut-off BELOW the cross-over frequency
(again chosen to be ω0 is unity), and the polar plot (e) is “inverted”.
102
Active first order filters
The maximum “gain” of the passive filters shown above is unity. However,
we can see how to engineer basic filter circuits with gain if we consider the
inverting op-amp circuit in two variations.
which has a transfer function:
( )
+
−=
+
−=21
2
1
2
2
111
CRjRR
RCRj
R
jGω
ωω
which is identical to the low-pass filter form shown above, but now multiplied
by a “gain” of –R2/R1 .
and
103
which has a transfer function:
( ) ( )
+
−=+
−=+
−=CRj
CRjRR
CjRCjR
CjRRjG
2
2
1
2
1
2
1
2
11/1 ωω
ωω
ωω
which is identical to the high-pass filter form shown above, but now
multiplied by a “gain” of –R2/R1 .
Second order filters
Whilst the filter ideas introduced above are useful, they also have some
limitations. Only low-pass and high-pass functions can be implemented and
the cut-off is limited to 20dB per decade. What if we would like steeper cut-
offs etc.? One basic answer is simple, we can chain filters together. For
example if we connect two low-pass filters together (putting a buffer between
them) we have:
104
This gives a transfer function equivalent to the first order low pass filter
multiplied by itself, i.e:
( )( ) ( )2
022
/11
11
11
11
ωωωωωω
jTjRCjRCjjG
+≡
+≡
+×
+=
which for high frequencies becomes ω02/ ω2 . Therefore at high frequencies
the magnitude is inversely proportional to ω2, whereas for the first order filter
it was inversely proportional to ω . Hence, for this second-order filter the
drop-off at high frequencies is 40dB per decade.
For the first order low-pass filter the transfer function is always of the form:
( )0/1
11
1ωωω
ωjTj
jG+
≡+
= .
However, does the second order low-pass filter always have a transfer
function of the form determined above? The answer is yes and no! This can
be illustrated by determining the response if we had not included the buffer.
The circuit is now:
105
To determine the transfer function of this we could use mesh/loop or node
analysis – a very similar problem was on a P2 tutorial sheet. The resulting
transfer function is:
( ) 20
20
222222 //311
311
311
ωωωωωωωωω
−+≡
−+≡
−+=
jTTjCRRCjjG
which is not identical in form to the transfer function determined above, but
has a similar overall structure. Note that it is the middle term which has
changed, and in general we can write an expression for a second-order low-
pass filter which is of the form:
( ) 20
20
2 //211
ωωωωζω
−+=
jjG
where ζ is the referred to as the damping. In the buffered case above ζ=1
and in the un-buffered case ζ=1.5 .
In general, we can conveniently design a second-order low-pass filter where
the damping ratio ζ and characteristic frequency ω0 can be controlled, by
using a Sallen-Key topology. Such a circuit is illustrated below:
106
For this circuit the transfer function is:
( ) ( ) 21212
2122 1
1RRCCRRCj
jGωω
ω−++
=
so,
( )2121
212
21210 2
and 1RRCC
RRCRRCC
+== ζω
and given that there are four components and only two parameters to set we
have enough freedom to independently control the damping ratio ζ and
characteristic frequency ω0 .
To illustrate the effect of varying the damping parameter a series of plots of
|G| for a second order low pass filter, where we have fixed ω0=1, but
changed the damping. Taking the form used above:
107
( )
( )( ) 2
02222
02
2
20
20
2
/4/1
1
//211
ωωζωωω
ωωωωζω
+−=
⇒−+
=
jG
jjG
0.000001
0.00001
0.0001
0.001
0.01
0.1
1
10
0.001 0.01 0.1 1 10 100 1000
|G(w
)|
Ang-frequency (w)
damp=0.1damp=1
damp=3
For any damping we can see that when ω<<ω0 (in this case ω<<1) then G=1
and when ω>>ω0 (in this case ω>>1) G=ω02/ ω2 , i.e. the roll-off is 40dB per
decade. However, the behaviour around ω=ω0 (in this case ω=1) varies
substantially.
For large values of damping (greater than 1) the transfer function for the
second order low pass filter can be factorised. For example, when the
damping is 10 we have:
( )00
20
20
2 /05.011
/95.1911
//2011
ωωωωωωωωω
jjjjG
×+×
×+≅
−+=
108
So the second order filter behaves as the “product” of two first order filters,
with cross-over frequencies of ω=ω0/19.95 and ω=ω0/0.05. This is evident
from the diagram above, where when the damping is 10 we can see two
cross over frequencies, the first at 1/19.95 and the second at 1/0.05, where
between them we see a first-order roll-off of 20dB per decade.
As the damping ratio is reduced the two cross-over frequencies come
together, until when the damping is unity we reach the point where the
behaviour is equivalent to two “cascaded” (and buffered) identical first-order
filters as shown earlier. That is, when damping is unity:
( )00
20
20
2 /11
/11
//211
ωωωωωωωωω
jjjjG
+×
+=
−+=
If the damping is further reduced (less than unity) then it is less convenient to
factorise into two first order terms. The factors in front of the jω terms
become complex! The resulting behaviour can be seen in the diagram
above. For example, when the damping is 0.1, we see a resonance around
the cross-over frequency.
It is easier to understand the relevance of the above to practical filter usage
is we introduce an alternative way of characterising the low-pass filter. The
cross over frequency is a key characteristic, but if we observe the behaviour
in the responses shown above then we can see that the roll-off “appears” to
begin at different points for different damping values. This can be allowed for
by adjusting the cross-over frequencies so that the roll-off occurs at a similar
point in each case. To do this we must select a point to characterise the
109
“start” of the roll-off. The point normally chosen is the -3dB point. That is
where the gain of the low-pass filter has dropped by 3dB from its low
frequency value. This is also where the amplitude of the transfer function has
dropped by a factor of 2 (because ( ) dB32/1log20 10 −= ). Adjusting the
values of the cross-over frequencies in this way for a range of damping we
get the plots shown below:
0.000001
0.00001
0.0001
0.001
0.01
0.1
1
10
0.001 0.01 0.1 1 10 100 1000
|G(w
)|
Ang-frequency (w)
damp=0.1
damp=0.707damp=1
We can now see that for a fixed 3dB point, reducing the damping in a
second order low pass filter causes the transfer function to roll-off earlier
(qualitatively, this is sometimes referred to as “sharpening” the response).
The cost of this is the potential introduction of a peak around the cross-over
frequency. (Note: ω0 now varies from curve-to-curve, and is not always =1!)
It turns out that the “best” which can be done without introducing any peak is
when the damping is equal to 707.02/1 ≅ . (This has the advantage that
ω0 is the 3dB frequency – which avoids any possible confusion when
110
describing the filter) In this case the second-order low pass transfer function
becomes:
( )20
20
2 //211
ωωωωω
−+=
jjG
and the amplitude of the transfer function is given by:
( ) 40
42 /11
ωωω
+=jG
This is referred to as a second-order low-pass Butterworth filter, named after
the British engineer Stephen Butterworth (~1930).
A high-pass second order filter can be constructed in a similar way to that
outlined above, but swapping the resistors and capacitors, to give:
For this circuit the transfer function is:
( ) ( ) 21212
211
21212
2 1 RRCCCCRjRRCCjG
ωωωω
−++−
=
so with,
111
( )2121
211
21210 2
and 1RRCC
CCRRRCC
+== ζω
we have:
( ) 20
20
20
2
2 //21/
ωωωωζωωω−+
−=
jjG
and the amplitude of the transfer function is then:
( )( ) 2
02222
02
20
2
2/4/1
/
ωωζωω
ωωω+−
=jG
This leads to amplitude Bode plots of the form:
0.000001
0.00001
0.0001
0.001
0.01
0.1
1
10
0.001 0.01 0.1 1 10 100 1000
|G(w
)|
Ang-frequency (w)
damp=0.1
damp=0.707damp=1
Or, if the cross-over frequencies are adjusted to make the -3dB points
identical for each case we have:
112
0.000001
0.00001
0.0001
0.001
0.01
0.1
1
10
0.001 0.01 0.1 1 10 100 1000
|G(w
)|
Ang-frequency (w)
damp=0.1
damp=0.707damp=1
Band pass A further form of second-order filter is possible. A good way to see this
concept is to re-consider the first order low-pass filter introduced earlier.
Remember the basic first order low pass filter consists of a resistor and
capacitor:
which resulted in a transfer function between the input and output of the
form:
113
( )0/1
11
11
1ωωωω
ωjTjRCj
jG+
≡+
≡+
=
Now, consider what happens if an inductor is placed in parallel with the
capacitor, to give:
This changes the impedance of the reactive part from
Cjω1
to
11
−
+
LjCj
ωω
or, if we consider admittances it changes
Cjω to LjCj
ωω 1
+
If we substitute this into the low pass filter transfer function we generate a
new transfer function:
( )
( ) ( )
( )BAB
B
BA
BAbp
lowlow
jj
jj
jjjG
jjG
ωωωωωωω
ωωωωω
ωωωωω
ωωω
//1/
/
//11
/11
2
2
_0
−+=
+−=
++=
→+
=
114
which is the transfer function for a band-pass filter, conventionally expressed
in the form:
( ) 20
20
0
//21/2
ωωωωζωωζω−+
=j
jjGbp
or, introducing the “quality factor” Q this can be written:
( )20
20
0
//11
/1
ωωωω
ωωω
−+=
jQ
jQjGbp
and the amplitude of the transfer function is:
( )( ) 2
02222
02
0
/4/1
/2
ωωζωω
ωζωω+−
=jGbp
We can again plot the amplitude Bode diagram for various damping, giving:
0.000001
0.00001
0.0001
0.001
0.01
0.1
10.001 0.01 0.1 1 10 100 1000
|G(w
)|
Ang-frequency (w)
damp=0.1
damp=0.707damp=1
In this case the curves are symmetric, and the cross-over (natural) frequency
is the useful characteristic. It is also clear that at ω=ω0 we have:
115
( )( )
1/4/1
/220
20
22
20
20
000 =
+−=
ωωζωω
ωζωωjGbp
The behaviour of this filter is also nicely illustrated if we plot |G| on a linear
scale, giving:
0
0.2
0.4
0.6
0.8
1
0.001 0.01 0.1 1 10 100 1000
|G(w
)|
Ang-frequency (w)
damp=0.1
damp=0.707damp=1
Clearly, the smaller the damping the narrower the range of frequencies that
are passed (centred around the natural frequency).
This is exactly what happens with the resistor-capacitor-inductor circuit
shown above. However, inductors can be inconvenient (they are both large
and imperfect), and it is often more useful to use active circuits based on
resistors and capacitors only (together with an op-amp). There are many
active implementations of the band pass filter. A simple implementation is:
116
Recalling the well-known transfer function when only the resistors are
present (i.e. 12 / RRG −= ) we can easily see that we now have:
( )
( )( )
( ) 22112
2211
2211
2211
21
22112
2211
21
11
22
2
1
1
11
RCRCRCRCjRCRCj
RCRCRC
RCRCRCRCjRCj
CjR
RCjR
jG
ωωω
ωωω
ω
ωω
−+++
+
−=
−++−=
+
+
−=
which is of the form introduced above for a second-order band-pass filter,
but now with a gain (or loss) at the cross-over (natural) frequency given by
the term in front.
117
General second order filter implementation.
In the P2 course you were introduced to a multiple feedback circuit design
which could be used to implement second-order low-pass, band-pass and
high-pass filters. The circuit is reproduced below:
Figure 40: Generalised multiple feedback circuit.
Where the “Y”s represent the admittances of the various feedback
components, i.e. Yn=1/Zn , so for a resistor YR=1/R and for a capacitor
YC=jωC. The analysis of the circuit was presented in the P2 course (if you
are not confident you could analyse it if asked to do so then remind yourself
of this because it is important!). The resulting transfer function is:
( ) 4343215
310
YYYYYYYYY
vvG
i ++++−
==
118
Now, by comparing this transfer function with the general transfer functions
for the various filter types we can see that by selecting appropriate
components they can be implemented. For example, if we use:
11 /1 RY =
22 /1 RY =
33 CjY ω=
44 CjY ω=
55 /1 RY =
we get the transfer function:
( )( )
( )
( )
( )
+
−+
+
+
+
+
+
−=
+
−+
+
+
+
−=
21
21543
243
21
21
4321
21
43
3
1
5
21
21543
243
21
21
21
21
1
53
1
1
RRRRRCCCC
RRRRj
CCRR
RRj
CCC
RR
RRRRRCCCC
RRRRj
RRRR
RRCj
jGbp
ωω
ω
ωω
ωω
which is identical in form to the general second-order band-pass filter
transfer function introduced above.
Alternative component selection allows low and high pass to be
implemented.
119
Band-stop filter The band-stop filter (sometimes called a notch filter) is intended to remove
signals around a narrow frequency range. For example we mentioned earlier
that 50Hz mains interference is a common problem, so a 50Hz “notch” filter
might be useful. A simple implementation is to feed a signal though both a
second order low-pass filter and a second order high-pass filter
simultaneously and then to combine the signals with a further op-amp.
Consider the following circuit:
The transfer function for the circuit around the upper left op amp, between
Vin and Vlp is given by:
( ) 20
20 //21
1ωωωωζ
ω−+
=j
jGlp
The transfer function for the circuit around the lower left op amp, between Vin
and Vhp is given by:
120
( ) 20
20
20
2
//21/
ωωωωζωωω−+
−=
jjGhp
The circuit around the right hand op amp is then an inverting summing
amplifier, so the overall transfer function between Vin and Vout is:
( ) 20
20
20
2
//211/
ωωωωζωωω
−+−
=j
jGbs
If we plot the modulus of this for various damping ratios (where we have
fixed the cross-over (natural) frequency to be unity) we have:
0
0.2
0.4
0.6
0.8
1
1.2
0.001 0.01 0.1 1 10 100 1000
|G(w
)|
Ang-frequency (w)
damp=0.1damp=0.707damp=1
Higher order filters
In the above material we have discussed first-order filters (low-pass and
high-pass) and second order filters (low-pass, band-pass, high-pass and
stop-band). Clearly it is possible to engineer “higher” order filters, which
generally have steeper characteristics. For example, the first-order low-pass
filter had a roll-off of 20dB per decade and the second order low-pass filter
121
had a roll-off of 40dB per decade. Clearly third, fourth, etc. order low-pass
filters would have roll-offs of 60, 80, etc. dB per decade. These are created
by connecting the correct number of first and second order filters in series.
Applications of filters Noise rejection Filters are often used to limit the noise bandwidth (see definition below) of
the input signal in order to reduce the amount of noise in the input signal.
The quality of a signal in the presence of noise is most often specified by the
signal-to-noise ratio (SNR) which is
≡
=
n
s
n
s
VV
VVSNR 102
2
10 log20log10
where is the rms value of the signal and is the rms value of the noise.
This definition shows that a signal-to-noise ratio of one means that the noise power equals the signal power. Initially, it would appear that under these conditions it would be difficult to distinguish a signal from noise to obtain an accurate measurement. The signal in figure (31) shows that as expected with a SNR of one it is difficult to make an estimate of the signal component over a short interval, in this case less
than 0.1s. However, this figure also shows that over a longer timescale it can be possible to distinguish a periodic signal from the random signal caused by noise. In effect you can recognise the underlying 10 Hz signal
122
using its predictable periodic behaviour. In an instrumentation system the
same effect can be achieved by filtering this voltage to reduce the amount of
noise.
Figure 31: A signal at 10Hz with noise when the signal-to-noise ratio is one.
Signal Bandwidth and Noise Bandwidth
One of the useful functions of a low-pass filter is that it limits the noise
bandwidth of the signal.
For example, the transfer function of a simple RC low-pass filter, with a
characteristic frequency, , can be written in the form
123
Figure 37: The frequency response of a simple RC low-pass filter.
This is a smooth transfer function and it has to be assumed that some attenuation of the frequencies of interest is acceptable. By convention it is usually assumed that this filter will be used for signal frequencies
less than . This means that it is assumed that it is acceptable to attenuate some signal frequencies by as much as 3dB and by
convention the signal bandwidth of this filter is .
The signal bandwidth of the low-pass filter is only determined by convention,
and, in practice it is determined by the amount of acceptable signal
attenuation in a particular application. In contrast, the noise bandwidth unambiguously arises from a calculation of the noise at the output of the filter.
124
For a white noise source the noise power in a small frequency range df is
independent of the actual centre frequency. As with independent noise
sources the total noise arising from contributions in different frequency
ranges is determined by adding the power in each frequency band. Hence
for a white noise source with a noise power of per Hertz, the output
power from the filter will be
which can be integrated to give
This is equivalent to the output of an ideal low-pass filter with a bandwidth of
1.57 . This is the noise equivalent band-width of the filter. Using the characteristic frequency of the filter to determine the noise bandwidth of the filter would therefore significantly under-estimate the amount of noise that would be present at the output of the filter. However, not
unexpectedly, for filters with faster roll-offs at higher frequencies the noise
equivalent bandwidth quickly approaches the characteristic frequency of the
filter.
Anti-aliasing
Aliasing is a problem that arises when a signal that is continuously
varying in time is sampled in time. If a high frequency signal is
sampled too infrequently then once it is sampled it appears to be a
signal at a different (lower) frequency. To avoid aliasing problems the
125
sampling frequency must be twice the maximum frequency of the
input signal. This is the Nyquist (or Nyquist–Shannon) sampling theorem.
Increasing the sampling rate of a signal will prevent aliasing problems for
signal frequencies, but, it can never prevent aliasing of high frequency noise.
Analogue anti-aliasing filters are therefore included in all well designed
instrumentation systems. However, good analogue filters, with a sharp roll-
off, requires a large number of components and must be carefully designed.
Even then they are not ideal. Once they have been sampled signal are
therefore sometimes filtered by the digital processor.
126
AC SIGNALS IN INSTRUMENTATION – the lock-in amplifier Introduction In the earlier description of a bridge circuit it was assumed that a d.c. bias
voltage was applied to the elements of the bridge. However, there are some situations in which an a.c. bias voltage is either necessary or beneficial. One example of a situation in which an a.c. bias voltage is necessary is when taking measurements from a bridge circuit formed by capacitors. The problem with using a d.c. bias voltage with this type of
bridge circuit is that the d.c. impedance of all capacitors is infinite and it is
impossible to detect a change in capacitance. In this case to reduce the
impedance of the capacitors an a.c. bias voltage should be applied to the
bridge circuit.
An example of a situation in which an a.c. bias voltage is beneficial is when there is a small signal expected near 50 Hz. Since this is the same frequency as the mains power supply, it is particularly vulnerable to interference. An a.c. bias voltage can then be used to shift the sensor output signal to a different frequency to separate the signal frequency of interest from the frequency of the interfering signal. Filters can then be used to reject the interference whilst amplifying
the signal. To support these two critical functions an important system
component is required – the lock-in amplifier. The lock-in amplifier removes
the effects of the a.c. bias from the filtered and amplified output of the
bridge circuit. It also acts as a narrow pass-band filter around the operating
frequency.
127
Bridge Circuits and Capacitance Based Sensors
A change of resistance is only one of the means that can be used to create
a circuit element that is sensitive to a physical variable. The other two
properties that could be used are capacitance and inductance. An example
of the use of inductance to measure a variable, is the use of a moving core
within a inductor to detect displacement. In addition, changes in
capacitance arising from variations in dielectric constant caused by
absorbed gases can also be detected. This mechanism then forms the
basis of gas detectors, including detectors for humidity and explosive or
poisonous gases.
One approach to converting changes in inductance or capacitance to a voltage change is to simply form a bridge circuit containing either inductors or capacitors. However, the low-frequency impedance of an
inductance is very small and its dc impedance is zero. A real dc power
supply, with a small but finite output impedance, would be unable to sustain
a voltage across this type of bridge. In contrast the impedance of a
capacitor at low frequencies is very large. It would therefore be easy to
sustain a dc voltage across a bridge circuit containing capacitors. However,
this large impedance will make it impossible to sense the voltage across
the bridge circuit without altering its value.
The problems with dc bridges containing inductors or capacitors can only be overcome if the impedance of the inductors can be increased and the impedance of capacitances can be decreased. Both these
128
objectives can be achieved using the same approach: applying an ac voltage of a known frequency to the bridge circuit. Output Signal Processing
Figure 54: Bridge circuit formed with capacitors biased by
an ac signal generator.
To understand the effects of applying an a.c. bias voltage to a bridge circuit
consider a bridge formed from two sensor capacitors and two reference capacitors as shown in figure (54). If the voltage applied to
this bridge circuit is , then the differential output voltage is
Now assume that the value of the sensing capacitor has been chosen so that
129
and the modulation of the capacitance value caused by the physical
variable, , is only a small fraction of its average value so that
, then
The final stage of the analysis is to assume that the physical stimulus has a single dominant frequency so that
where ωm is the frequency of the physical effect itself, and that the
bias voltage has the form
where ω r is the frequency of the applied bias voltage.
Then the output signal from the bridge circuit is:
Assuming that (reference/bias frequency is greater than the physical-effect/signal frequency) this means that the output signal
contains two frequencies centred on separated by .
130
With recent and continuing improvements in ADC performance it may be
feasible to digitise this output signal before extracting the required
information in a digital processor. However, in order to obtain a reasonable
impedance in the bridge circuit it may be necessary to operate at
frequencies of more than 100KHz. This could preclude the use of the
cheaper successive approximation ADCs. Even in situations in which it
may be possible to directly digitise the ac signal, it might therefore be
cheaper to process the signal to reduce its frequency without losing
relevant information. The sampling rate of the ADC will then be determined
by the rate of change of the physical variable, , rather than the higher
frequency of the applied ac stimulus, . An instrumentation system that employs an ac bridge may therefore require a technique to remove the effects of the a.c. signal that has been applied to a bridge circuit.
One approach to removing the applied ac signal is to use the circuit shown in figure (55). This circuit relies upon the existence of a reference
signal, r(t), which can be used to switch the input signal, s(t), through one
of two alternate paths. These two paths are designed so that one path acts
as an amplifier with a gain of +1, whilst the other path acts as an inverting
amplifier with a gain of -1. In effect this switching between two paths
multiplies the input signal by a square wave with an amplitude of 1± .
131
Figure 55: Block diagram of a lock-in amplifier.
To analyse the system assume that the reference signal r(t) is a square
wave of frequency (where is the frequency of the a.c. signal
generator). By determining the position of an electronic switch between A
and B this signal effectively multiplies the signal s(t) by a square wave. The
Fourier series of this type of square wave is (see HLT):
This is then multiplied by the input signal
and hence the output from the switch multiplier is
132
The low-pass filter which follows the switch is then designed so that
its cut-off frequency is significantly less than . Hence the output from the low-pass filter is:
where is the magnitude response of the LPF at frequency
and since
The combination of a multiplier and a low-pass filter that respond to a
narrow range of frequencies is a commonly used component of an
instrumentation system. It is often referred to as a lock-in amplifier.
Interference and Noise Reduction
Figure 56: A bridge circuit of resistors with an ac input voltage.
The use of a lock-in amplifier has been introduced in the context of either all-capacitance or all-inductance bridge circuits. In these
133
situations the bridge circuit is biased with an a.c. signal in order to control
the impedance of the elements within the bridge. The lock-in amplifier is
then used to reduce the frequency at which the output signal has to be
sampled without losing any information about the physical process being
monitored. However, lock-in amplifiers can also be used in other systems in order to avoid strong sources of interference or noise.
To understand how a lock-in amplifier can be used to avoid interference or noise consider a bridge formed from two sensor resistors and two reference resistors. If the voltage applied to this bridge
circuit is , then the differential output voltage is
Now assume that the value of the sensitive resistor has been chosen so
that
and the modulation of the resistance value caused by the physical variable,
, is only a small fraction of its average value so that , then
Problems arise when the system has to be designed to detect small
changes in resistance. In this situation small changes in the output signal
will be vulnerable to either external interference or noise arising within
subsequent analogue circuits. Some protection from interference can be obtained by filtering the signal. However, any filters must be designed to
pass frequencies that contain information about the physical variable that is
134
being measured. It can then be very difficult to obtain a filter with a sharp enough cut-off to reject a strong source of interference and in the worst-case situation the interfering signal may occur within the interesting frequency range. Analogue filtering is then impossible and there is a risk that the interfering signal could sometimes be large enough to cause saturation of the input signal to the ADC.
The solution in these situations is to apply an a.c. bias voltage to the bridge circuit
To understand the effect of this consider a situation in which the physical
stimulus has a single dominant frequency, , so that
In this situation the output signal from the bridge circuit is:
This can be rewritten in the form
to show that the output now contains two signals at frequencies
and .
135
Figure 57: A schematic diagram of the component parts required to shift
the intermediate signal frequency in order to avoid noise and or
interference. Once all the analogue amplification and filtering has been
performed in a frequency range chosen by the designer a lock-in amplifier
can be used to reduce the output frequency (Note that the two points
labelled A are connected together as are the points labelled B)
Applying an a.c. signal of known frequency to the bridge circuit therefore enables the designer to create an output signal with two well controlled frequency components. Although the amplitude of these
two components is small the designer can select the value to in order to
avoid frequencies at which interference is expected (usually the frequency
of the local mains and/or its harmonics). Furthermore, the value of can be chosen to shift the output frequencies so far from the frequency of
136
any interference that even a relatively simple band-pass filter can
amplify the signal at frequencies around whilst rejecting interference. Once the signal has been amplified without amplifying any
interference its frequency can be reduced using a switching multiplier and
low-pass filter prior to digitisation without fear of degradation.
Phase-Difference Detection or Phase-Sensitive Detection (not really part of this course, but rather interesting….)
Another approach to detecting a change in capacitance is to place a reference resistor in series with a variable sensor capacitance to create a low-pass filter. If the input voltage to this circuit is close to the 3dB frequency of the filter then variations in the capacitance values will cause a measurable change in the phase difference between the input signal to the filter and its output. This phase difference can then be detected using a combination of a switching multiplier and low-pass filter
To understand how this can occur consider a difference between the phase
of two signals consider the situation in which the transducer modulates the
phase of a signal with a known frequency. In this situation the input signal
can be expressed as
Once again assume that the reference signal r(t) is a square wave of
frequency and amplitude , equivalent to a Fourier series of:
137
The output of the switching multiplier is then given by the product r(t)s(t).
Thus:
The 3-dB cut-off frequency of the low-pass filter is chosen to be well
below . Hence the multiplier products at frequencies 2 , 4 , 6 , etc
are eliminated and the output of the Phase-Sensitive Detector contains only
the phase-sensitive d.c component:
(assuming that the d.c. gain of the LPF is 1) Hence using this circuit it is possible to detect any phase difference between the reference signal and input signal. The circuit is therefore often referred to as a phase sensitive detector.
138
Mini summary
Alternating current (a.c.) biasing signals can be used in conjunction with
bridge circuits containing inductors or capacitors to control the impedance
of the elements of the bridge circuit so that a measurable output signal can
be created.
Alternating current (a.c.) biasing signals can be used with bridge circuits of
resistors in order to shift the frequency of the output signal to a pre-
determined frequency range. Filter circuits can then be designed in the
analogue signal processing stage which can amplify the signal but reject
interference whose frequency would otherwise be too close to the signal
frequency of interest.
A lock-in amplifier can be used after amplification and filtering to remove
the effects of the a.c. bias. The maximum required ADC sampling rate is
then determined by the rate of change of the physical variable being
measured rather than the much higher a.c. frequency chosen by the
designer.
The key component of the lock-in amplifier is a circuit that can also be used
to detect the phase between two signals of the same frequency. It is
therefore often known as a phase sensitive detector.